Publications

Modified

22 June 2026

The full, up-to-date list of publications below is generated automatically from a single BibTeX file (assets/references.bib) — the same file you can keep in sync with your reference manager (Zotero, etc.). Open-access full text for most items is available through HAL.

assets/references.bib is the single source of truth. Running make pubs (which calls scripts/build_publications.py) regenerates _publications.md, included just below. Nothing here is executed at render time, so building the site needs no Python — see the project README.

166 entries, generated from assets/references.bib.

Books

2021

I-Han (Sharon) Hsiao, Shaghayegh (Sherry) Sahebi, François Bouchet & Jill-Jênn Vie (eds.) (2021). Proceedings of the 14th International Conference on Educational Data Mining.

2020

Marco Barzman, Mélanie Gerphagnon, Olivier Mora, Genevieve Aubin-Houzelstein, Alain Bénard, Caroline Martin, George-Louis Baron, François Bouchet, Juliette Dibie, Jean-Francois Gibrat, Simon Hodson, Evelyne Lhoste, Yann Moulier-Boutang, Sébastien Perrot, Fabrice Phung, Christian Pichot, Mehdi Siné & Thierry Venin (2020). La transition numérique dans la recherche et l'enseignement supérieur à l'horizon 2040. Quae.

La transition numérique crée des opportunités et des défis inédits pour l’enseignement supérieur et la recherche publique. Cette prospective envisage des évolutions possibles et contrastées du fonctionnement de la recherche, de l’apprentissage et des modes de partage des savoirs.

Business & Economics / Accounting / Financial

Book chapters

2022

Katia Oliver-Quelennec, François Bouchet, Thibault Carron & Claire Pinçon (2022). Evaluating the transposition of a Learning Analytics Dashboard Co-design Tangible Tool to a Digital Tool. Open and Inclusive Educational Practice in the Digital World. Springer International Publishing.

In-person sessions of participative design are commonly used in the field of learning analytics, but to reach students not always available on-site (e.g., during a pandemic), they have to be adapted to online-only context. Card-based tools are a common co-design method to collect users’ needs, but this tangible format limits data collection and usage. We propose here two steps: first to adapt an existing co-design card deck-based method for an online use and then to leverage the benefits of the digital format to trace data with a research focus to study the dynamics of collaboration. We also assess whether the previously adapted digital tool has the same impacts as the face-to-face original method. Beyond the case described here, this chapter aims at identifying key factors and points of attention identified in adapting a card-based co-design method into a digital version for designing learning dashboards, which elements can be traced in order to study collaboration and propose future improvement to the method. This digital adaptation was tested by university students in different contexts (n = 177). All groups have successfully designed a dashboard, and using the original evaluation scales, users have evaluated our digital tool as almost as suitable as the original method. Then, we compared the use of the digital adaptation online with this use in face-to-face sessions which were less successful. We conclude by showing how the added traces open new perspectives to understand collaboration through links between speech acts and collaboration profiles or defining adapted dashboard for students.

Cards Co-design Learning analytics dashboard Participatory design

2021

Camila Morais Canellas, François Bouchet, Thibaut Arribe & Vanda Luengo (2021). Learning Analytics Metamodel: Assessing the Benefits of the Publishing Chain's Approach. Computer Supported Education. Springer.

In this work, we propose a learning analytics implementation based on a model-driven engineering approach. It aims at assessing the benefits that could arise from such an implementation, when pedagogical resources are produced via publishing chains, that already use the same approach to produce documents. Previously, we have discussed these potential benefits from a more theoretical point of view. In the present work, we present a concrete implementation of a metamodel to integrate a learning analytics system closely linked to the knowledge of the semantics and structure of any document produced, natively. Finally, we present an initial evaluation of this metamodel by modelers and discuss the limits of this metamodel and the future changes required.

2020

Guy Merlin Mbatchou Nkwetchoua, François Bouchet, Thibault Carron & Philippe Pernelle (2020). Towards a Model of Learner-Directed Learning : an Approach based on the Co-Construction of the Learning Scenario by the Learner. Online Teaching and Learning in Higher Education. Springer, Cham.

To improve the learning process, the evolution of learner’s characteristics (cognitive, affective, prior knowledge, workflow, organization, …) must be taken into account during the personalization or adaptation. This requires generating several scenarios (a description of activities, their order and links in the learning sequence as well as the expected outcome for the learner) adapted to the identified profiles. We propose a model which aims at improving learners’ learning processes by giving them control over two key aspects: (1) the steps of the learning scenario to be followed: after each learning goal is completed, the learner chooses the next one among several possible ones (in terms of their current knowledge) while respecting pedagogical constraints (time and quality of the solutions produced according to satisfaction thresholds); (2) the assessment mode: the learner chooses a mode corresponding to their own goals in terms of mastery, while respecting the minimum thresholds set by the teacher. We assess our approach with learners in terms of (a) adequacy of the model with learners’ expectations and (b) usability of the system through self-report questionnaires and an analysis of the data collected over 16 learners who used an implementation of our system on the LMS (Learning Management System) in the context of a real computer course. The results reveal that our model is mostly well-received, that various scenarios are indeed followed by the learners and that the 3 assessment modes have been used by learners.

Eric G. Poitras, Reza Feyzi Behnagh & François Bouchet (2020). A Dimensionality Reduction Method for Time Series Analysis of Student Behavior to Predict Dropout in Massive Open Online Courses. Adoption of Data Analytics in Higher Education Learning and Teaching. Springer International Publishing.

Student attrition is one of the most frequently stated problems with massive open online courses. Although there is a growing body of research that investigates the factors leading to withdrawing from a course, there is also a need for data-driven solutions for early detection as a means to take remedial action. In this case study, we examine the use of several data preprocessing techniques to model attrition on the basis of students’ interactions with course materials and resources. Data for this study were obtained using the Open University Learning Analytics Dataset (OULAD), enabling the analysis of daily summaries of clickstream data. The data were segmented using a variable-sized overlapping window to take into account assignment submission dates as a context-sensitive factor for student attrition. In each sliding time window, features were extracted to characterize student interactions with curricular materials and converted to a set of linearly uncorrelated principal components. The analysis demonstrates that relatively accurate detection of the likelihood of students dropping out from a course can be attained within approximately 10 weeks from the course start date or before completion of assignments worth 20% of the final grade. Although a decision tree model outperformed alternative approaches to model student attrition, a log-linear model performed comparably well on the sparse representation of student interactions obtained through principal components analysis. We discuss the implications of these findings for early identification of at-risk students and the design of analytics dashboards for advisors and tutors.

Clickstream log data Massive open online course Principal component analysis Student attrition Time series classification

2016

John Ranellucci, Eric G. Poitras, François Bouchet, Susanne P. Lajoie & Nathan C. Hall (2016). Understanding Emotional Expressions in Social Media through Data Mining. Emotions, Technology, and Social Media. Academic Press.

2014

Jean-Paul Sansonnet & François Bouchet (2014). A Framework Covering the Influence of FFM/NEO PI-R Traits over the Dialogical Process of Rational Agents. Agents and Artificial Intelligence. Springer Berlin Heidelberg.

In this article we address coverage and comprehensiveness issues raised by the integration of a large class of psychological phenomena into rational dialogical agents. These two issues are handled through the definition of a generic framework based on the notion of personality engine, which makes it possible to reify in separate modules in one hand the application-dependent parts and on the other hand the resources involved in the representation of the psychological phenomena. We introduce an enriched taxonomy of personality traits, based on the well-used ffm/neo pi-r taxonomy and we show how it can be applied on an example of agents, taken from the literature. Then we introduce the necessary concepts for modeling a personality engine. A case study, using a simplified world of dialogical agents, shows how those agents can be provided with a personality engine affecting the way they communicate with each other, and demonstrates how it can be used to implement the example. Finally, we compare our approach to other attempts at implementing personality features in rational agents.

2013

Roger Azevedo, Jason M. Harley, Gregory J. Trevors, Melissa Duffy, Reza Feyzi-Behnagh, François Bouchet & Ronald S. Landis (2013). Using trace data to examine the complex roles of cognitive, metacognitive, and emotional self-regulatory processes during learning with multi-agents systems. International Handbook of Metacognition and Learning Technologies. Springer.

This chapter emphasizes the importance of using multi-channel trace data to examine the complex roles of cognitive, affective, and metacognitive (CAM) self-regulatory processes deployed by students during learning with multi-agent systems. We argue that tracing these processes as they unfold in real-time is key to understanding how they contribute both individually and together to learning and problem solving. In this chapter we describe MetaTutor (a multi-agent, intelligent hypermedia system) and how it can be used to facilitate learning of complex biological topics and as a research tool to examine the role of CAM processes used by learners. Following a description of the theoretical perspective and underlying assumptions of self-regulated learning (SRL) as an event, we provide empirical evidence from five different trace data, including concurrent think-alouds, eye-tracking, note taking and drawing, log-files, and facial recognition, to exemplify how these diverse sources of data help understand the complexity of CAM processes and their relation to learning. Lastly, we provide implications for future research of advanced leaning technologies (ALTs) that focus on examining the role of CAM processes during SRL with these powerful, yet challenging, technological environments.

2012

François Bouchet & Jean-Paul Sansonnet (2012). Intelligent Agents with Personality: from Adjectives to Behavioral Schemes. Cognitively Informed Interfaces: System Design and Development. IGI Global.

Conversational agents are a promising interface between humans and computers; but to be acceptable as the virtual humans they pretend to be, they need to be given one of the key elements used to define human beings: a personality. As existing personality taxonomies have been defined only for description, we present in this chapter a methodology dedicated to the definition of a computationally-oriented taxonomy, in order to use it to implement personality traits in conversational agents. First, a significant set of personality-traits adjectives is registered from thesaurus sources. Then, the lexical semantics related to personality-traits is extracted while using the WordNet database and it is given a formal representation in terms of so-called Behavioral Schemes. Finally, we propose a framework for the implementation of those schemes as influence operators controlling the decision process and the plan/action scheduling of a rational agent.

Journal articles

2024

Mélina Verger, François Bouchet, Sébastien Lallé & Vanda Luengo (2024). Intersectionality : deepen algorithmic fairness evaluation. The case study of academic performance prediction using data from online courses. STICEF (Sciences et Technologies de l'Information et de la Communication pour l'Éducation et la Formation), 31(1).

Assessing the algorithmic fairness of predictive models used in education has become cru-cial to ensure that, when deployed, they are not biased in favor or against certain learners. Until now, analyses have focused on assessing algorithmic fairness with regard to sensitive attributes in the data, such as gender, independently of each other. However, Crenshaw (1989)’s theory of inter-sectionality defends the idea that the influences of several sensitive attributes together produce unique and different discriminations for certain sub-groups of individuals. Thus, in this paper, we propose to extend (Verger, Bouchet et al., 2023)’s work with additional analyses of intersectional discriminations that are in the outcomes of predictive models. These models were used in the context of predicting success in online courses, using open educational data, specifically data from OULAD (Kuzilek et al., 2017). Our results shed light on algorithmic discriminations that were not identifiable from traditio-nal analyses, as well as determining the influence of each attribute on discriminations through their interactions with other attributes.

prediction équité algorithmique équité algorithmique intersectionnalité prédiction algorithmic fairness intersectionality prediction intersectionality intersectionnalité prédiction algorithmic fairness

Thomas Sergent, Morgane Daniel, François Bouchet & Thibault Carron (2024). How Can Self-Evaluation and Self-Efficacy Skills of Young Learners be Scaffolded in a Web Application?. IEEE Transactions on Learning Technologies, 17, pp. 1184–1197.

Self-regulated learning (SRL) skills are critical for students of all ages to maximize their learning. Two key processes of SRL are being aware of one's performance (self-evaluation) and believing in one's capabilities to produce given attainments (self-efficacy). To assess and improve these capabilities in young children (5 to 8), we use a literacy web-application where we introduced two randomly triggered prompts to evaluate perceived difficulty and desired difficulty. Comparing students' actual performance with their responses to self-regulatory prompts provides information about their ability to self-regulate their learning, in particular their self-evaluation and self-efficacy. The novelty of this work resides in: (a) studying the SRL of young children (5 to 8) in digital learning environments while learning another task (reading in our case) (b) measuring and improving some SRL abilities themselves and not only measuring and improving academic results in other tasks (c) the large number of students on which the studies were carried (over 400,000). Using 15,982,994 responses from 467,116 students, we first measured two types of SRL deficits, and then we assessed how a scaffolding and remediation strategy can reduce these deficits. In Study 1, we compare a group receiving remediation feedback to a control group, whereas in Study 2, we determine the impact of age and level on the remediation efficiency. Our contribution is twofold (a) a method to address on the long term a deficit in self-evaluation or in self-efficacy in a digital learning environment (b) a corroboration of the fact that students who are academically at risk lack self-efficacy and avoid tackling challenging exercises compared to their level. We therefore recommend that digital learning environments integrate an overlay of self-regulated learning such as self-evaluation and self-efficacy remediation loops, especially for younger students and students that are struggling academically. We included notes for educational practitioners at the end of this article for this purpose.

Mélina Verger, Chunyang Fan, Sébastien Lallé, François Bouchet & Vanda Luengo (2024). A Comprehensive Study on Evaluating and Mitigating Algorithmic Unfairness with the MADD Metric. Journal of Educational Data Mining, 16(1), pp. 365–409.

Predictive student models are increasingly used in learning environments due to their ability to enhance educational outcomes and support stakeholders in making informed decisions. However, predictive models can be biased and produce unfair outcomes, leading to potential discrimination against certain individuals and harmful long-term implications. This has prompted research on fairness metrics meant to capture and quantify such biases. Nonetheless, current metrics primarily focus on predictive performance comparisons between groups, without considering the behavior of the models or the severity of the biases in the outcomes. To address this gap, we proposed a novel metric in a previous work (Verger et al., 2023) named Model Absolute Density Distance (MADD), measuring algorithmic unfairness as the difference of the probability distributions of the model’s outcomes. In this paper, we extended our previous work with two major additions. Firstly, we provided theoretical and practical considerations on a hyperparameter of MADD, named bandwidth, useful for optimal measurement of fairness with this metric. Secondly, we demonstrated how MADD can be used not only to measure unfairness but also to mitigate it through postprocessing of the model’s outcomes while preserving its accuracy. We experimented with our approach on the same task of predicting student success in online courses as our previous work, and obtained successful results. To facilitate replication and future usages of MADD in different contexts, we developed an open-source Python package called maddlib (https://pypi.org/project/maddlib/). Altogether, our work contributes to advancing the research on fair student models in education.

classification fairness metric models' behaviors sensitive features student modeling unfairness mitigation

2022

Roger Azevedo, François Bouchet, Melissa Duffy, Jason M. Harley, Michelle Taub, Gregory J. Trevors, Elizabeth Cloude, Daryn Dever, Megan Wiedbusch, Franz Wortha & Rebeca Cerezo (2022). Lessons Learned and Future Directions of MetaTutor: Leveraging Multichannel Data to Scaffold Self-Regulated Learning with an Intelligent Tutoring System. Frontiers in Psychology, 13.

Self-regulated learning (SRL) is critical to learning, reasoning, and problem solving across tasks, domains, and contexts. Despite its importance, research shows that not all learners are equally skilled in accurately and dynamically monitoring and regulating their self-regulatory processes. Therefore, learning technologies, such as intelligent tutoring systems (ITSs), have been designed to measure and foster SRL. This paper presents an overview of over 10 years of research on SRL with MetaTutor, a hypermedia-based ITS designed to scaffold college students’ SRL while they learn about the human circulatory system. MetaTutor’s architecture and instructional features are designed based on models of SRL, empirical evidence on human and computerized tutoring principles of multimedia learning, Artificial Intelligence (AI) in educational systems for metacognition and SRL, and research on SRL from our team and that of other researchers. We present MetaTutor followed by a synthesis of key research findings on the effectiveness of various versions of the system (e.g., adaptive scaffolding vs. no scaffolding of self-regulatory behavior) on learning outcomes. First, we focus on findings from self-reports, learning outcomes, and multimodal data (e.g., log files, eye tracking, facial expressions of emotion, screen recordings) and their contributions to our understanding of SRL with an ITS. Second, we elaborate on the role of embedded pedagogical agents as external regulators designed to scaffold learners’ cognitive and metacognitive SRL strategies. Third, we highlight and elaborate on the contributions of multimodal data in measuring and understanding the role of cognitive, affective, metacognitive, and motivational (CAMM) processes. Additionally, we unpack some of the challenges these data pose for designing real-time instructional interventions that scaffold SRL. Fourth, we present existing theoretical, methodological, and analytical challenges and briefly discuss lessons learned and open challenges.

intelligent tutoring systems learning metacognition multimodal data pedagogical agents scaffolding self-regulated learning trace data

Rawad Chaker, François Bouchet & Rémi Bachelet (2022). How do online learning intentions lead to learning outcomes? The mediating effect of the autotelic dimension of flow in a MOOC. Computers in Human Behavior, 134, pp. 107306.

We focus on the predictors of persistence and achievement in online learning by studying the students’ learning intentions and their psychological states during learning activities. Flow/autotelic experience is a powerful predictor of engagement in MOOCs and online learning in general and relates to the deep involvement and sense of absorption during learning activities. Both theory and empirical evidence propose that predictors of flow in an educational setting include the need for belonging to a group of learners. Using path analyses and structural equation modeling, we verify the causal links between social intentions, autotelic experience and MOOC learning outcomes such as final grade and dropout. Using the Online Learning Enrollment Intentions (OLEI) scale, we find that in total six OLEI items predict MOOC success and dropout, with flow as a mediating effect. In two models, we verify “Autotelic experience” as a mediator between enrollment intentions and MOOC final grade and dropout. Our results highlight socially driven intentions as major factors to be considered in online learning environments. We draw theoretical and practical implications for MOOC design, considering explicit communication about the provided learning environment and tools towards a socially shared learning experience.

Adult learning Distance education and online learning Dropout Enrollment intentions Flow MOOC

Thomas Sergent, Morgane Daniel, François Bouchet & Thibault Carron (2022). Détection de déficits d'auto-évaluation et d'auto-efficacité et remédiation dans un EIAH. 29(2).

Des travaux de recherche montrent que la capacité à autoréguler son apprentissage a un impact significatif positif sur les résultats scolaires. Nous présentons ici une étude visant à détecter des déficits d'autorégulation de l'apprentissage pour de jeunes élèves, dans le contexte d'une application web d'apprentissage de la lecture. À partir des réponses de 467 116 élèves à deux questions évaluant la difficulté perçue et la difficulté voulue, nous proposons une définition opérationnelle de différentes formes de déficits et mesurons ensuite l'impact de deux stratégies de remédiation pour les réduire. Les résultats soulignent la possibilité d'étayer les compétences d'apprentissage autorégulé dans une application Web dès le plus jeune âge, tout en apprenant une autre compétence.

2021

Marco Barzman, Mélanie Gerphagnon, Geneviève Aubin-Houzelstein, Georges-Louis Baron, Alain Benard, François Bouchet, Juliette Dibie-Barthelemy, Jean-Francois Gibrat, Simon Hodson, Evelyne Lhoste, Caroline Martin, Yann Moulier-Boutang, Sébastien Perrot, Fabrice Phung, Christian Pichot, Mehdi Siné, Thierry Venin & Olivier Mora (2021). Exploring Digital Transformation in Higher Education and Research via Scenarios. Journal of Futures Studies, 25(3), pp. 65–78.

Digital transformation induces rapid and profound changes in higher education and research (HER). With this foresight, INRAE and Agreenium, two French public HER institutions centered on food, agriculture and the environment, explore the challenges they face in a world increasingly dependent on digital resources. Four scenarios generated via morphological analysis provide researchers, teachers and institutions a heuristic framework to anticipate risks and opportunities in terms of: platformization of research, commodification of knowledge and the ascendency of data; pressing demands to respond to planet-wide emergencies; renewed multi-stakeholder relations between civil society and the academic community; and limits on energy and raw materials devoted to digital uses.

2020

Fatima Harrak, François Bouchet & Vanda Luengo (2020). Liens entre performance, assiduité et questions posées et votées en ligne dans le cadre d'une classe inversée. Sciences et Technologies de l'Information et de la Communication pour l'Éducation et la Formation, 27(2).

Les questions des élèves sont utiles pour leur apprentissage et pour l'adaptation pédagogique des enseignants. Nous étudions ici la nature des questions posées en ligne par les étudiants et comment le vote sur ces questions peut être lié à l’apprentissage. Nous avons donc développé un schéma de codage, puis conçu un annotateur automatique que nous avons appliqué à l'ensemble du corpus. Le résultat révèle que les votants réussissent mieux et assistent plus souvent au cours, mais le fait de poser des questions est associé à un apprentissage plus important.

2019

Fatima Harrak, François Bouchet & Vanda Luengo (2019). From Student Questions to Student Profiles in a Blended Learning Environment. Journal of Learning Analytics, 6(1), pp. 54–84.

The analysis of student questions can be used to improve the learning experience for both students and teachers. We investigated questions (N = 6457) asked before the class by first-year medicine/pharmacy students on an online platform, used by professors to prepare for Q&A sessions. Our long-term objectives are to help professors in categorizing those questions, and to provide students with feedback on the quality of their questions. To do so, we developed a coding scheme and then used it for automatic annotation of the whole corpus. We identified student characteristics from the typology of questions they asked using the k-means algorithm over four courses. Students were clustered based on question dimensions only. Then, we characterized the clusters by attributes not used for clustering, such as student grade, attendance, and number and popularity of questions asked. Two similar clusters always appeared (lower than average students with popular questions, and higher than average students with unpopular questions). We replicated these analyses on the same courses across different years to show the possibility of predicting student profiles online. This work shows the usefulness and validity of our coding scheme and the relevance of this approach to identify different student profiles.

2018

Jason M. Harley, Michelle Taub, Roger Azevedo & François Bouchet (2018). "Let’s set up some subgoals": Understanding human-pedagogical agent collaborations and their implications for learning and prompt and feedback compliance. IEEE Transactions on Learning Technologies, 11(1), pp. 54–66.

Research on collaborative learning between humans and virtual pedagogical agents represents a necessary extension to recent research on the conceptual, theoretical, methodological, analytical, and educational issues behind co- and socially-shared regulated learning between humans. This study presents a novel coding framework that was developed and used to describe collaborations between learners and a pedagogical agent (PA) during a subgoal setting activity with MetaTutor, an intelligent tutoring system. Learner-PA interactions were examined across two scaffolding conditions: prompt and feedback (PF), and control. Learners’ compliance to follow the PA’s prompts and feedback in the PF condition were also examined. Results demonstrated that learners followed the PA’s prompts and feedback to help them set more appropriate subgoals for their learning session the majority of the time. Descriptive statistics revealed that when subgoals were set collaboratively between learners and the PA, they generally lead to higher proportional learning gains when compared to less collaboratively set goals. Taken together, the results provide preliminary evidence that learners are both willing to engage in and benefit from collaborative interactions with PAs when immediate, directional feedback and the opportunity to try again are provided. Implications and future directions for extending co- and socially-shared regulated learning theories to include learner-PA interactions are proposed.

2016

Gregory Trevors, Reza Feyzi-Behnagh, Roger Azevedo & François Bouchet (2016). Self-Regulated Learning Processes Vary as a Function of Epistemic Beliefs and Contexts: Mixed Method Evidence from Eye Tracking and Concurrent and Retrospective Reports. Learning and Instruction, 42, pp. 31–46.

The objective of the current studies was to investigate how epistemic cognition related to specific phases and components of self-regulated learning and its adaptation to learning conditions of varying quality. In a multi-study, mixed method design, we presented university students with science content that relayed conceptual discrepancies and collected quantitative and qualitative data to study how students responded to discrepancies. In Study 1 (n = 42), we collected eye tracking patterns, study times, and metacognitive ratings and found that participants adapted their behavioral processing as a function of their epistemic cognition and discrepancy type. In Study 2 (n = 20), we collected concurrent think-aloud protocols and retrospective interviews to further explore why discrepancies were noticed (or not) and how they were resolved. Results revealed that prior knowledge and epistemic self-efficacy in oneself as an evaluator of knowledge emerged as important themes to detecting and efficiently resolving discrepancies. We conclude with a discussion of theoretical and methodological implications.

Jason M. Harley, Cassia K. Carter, Niki Papaioannou, François Bouchet, Ronald S. Landis, Roger Azevedo & Lana Karabachian (2016). Examining the predictive relationship between personality and emotion traits and students' agent-directed emotions: towards emotionally-adaptive agent-based learning environments. User Modeling and User-Adapted Interaction, 26(2), pp. 177–219.

2015

Jason M. Harley, François Bouchet, M. Sazzad Hussain, Roger Azevedo & Rafael A. Calvo (2015). A multi-componential analysis of emotions during complex learning with an intelligent multi-agent system. Computers in Human Behavior, 48, pp. 615–625.

This paper presents the evaluation of the synchronization of three emotional measurement methods (automatic facial expression recognition, self-report, electrodermal activity) and their agreement regarding learners' emotions. Data were collected from 67 undergraduates enrolled at a North American university whom learned about a complex science topic while interacting with MetaTutor, a multi-agent computerized learning environment. Videos of learners' facial expressions captured with a webcam were analyzed using automatic facial recognition software (FaceReader 5.0). Learners' physiological arousal was recorded using Affectiva's Q-Sensor 2.0 electrodermal activity measurement bracelet. Learners' self-reported their experience of 19 different emotional states on five different occasions during the learning session, which were used as markers to synchronize data from FaceReader and Q-Sensor. We found a high agreement between the facial and self-report data (75.6%), but low levels of agreement between them and the Q-Sensor data, suggesting that a tightly coupled relationship does not always exist between emotional response components.

2014

Michelle Taub, Roger Azevedo, François Bouchet & Babak Khosravifar (2014). Can the use of cognitive and metacognitive self-regulated learning strategies be predicted by learners’ levels of prior knowledge in hypermedia-learning environments?. Computers in Human Behavior, 39, pp. 356–367.

Research on self-regulated learning (SRL) in hypermedia-learning environments is a growing area of interest, and prior knowledge can influence how students interact with these systems. Fifty-two undergraduate students’ interactions with MetaTutor, a multiagent, hypermedia-based learning environment, were investigated, including how prior knowledge affected their use of SRL strategies. We expected that students with high prior knowledge would engage in significantly more cognitive and metacognitive SRL strategies, engage in different sequences of SRL strategies, spend more time engaging in SRL processes, and visit more pages that were relevant to their sub-goals than students with low prior knowledge. Results showed significant differences in the total use of SRL strategies between prior knowledge groups, and more specifically, revealed significant differences in the total use of metacognitive strategies, but not total cognitive strategies between prior knowledge groups. Results also revealed different sequences of use of SRL strategies between prior knowledge groups, and that students spent different amounts of time engaging in SRL processes; however, all students visited similar numbers of relevant pages. These results have important implications on designing multiagent, hypermedia environments; we can design pedagogical agents that adapt to students’ learning needs, based on their prior knowledge levels.

2013

François Bouchet & Jean-Paul Sansonnet (2013). Influence of FFM/NEO PI-R personality traits on the rational process of autonomous agents. Web Intelligence and Agent Systems, 11(3), pp. 203–220.

In this paper, we present an approach based on the principle that psychological capacities, especially personality traits, influence the decision making process of rational agents. Using a three-level (trait, facet, scheme) extension of the FFM/NEO PI-R taxonomy facilitating its computational implementation, we propose a model for the expression of personality traits in terms of so-called influence operators that add meta control rules to the cycle of rational BDI agents.We distinguish eight different classes of influence operators, depending on the step of the deliberation cycle that they influence and on the operand to which they are applied (goals, actions, plans or functions). Concrete examples are provided through a complete definition of the operators necessary to express one of the personality traits of the Five Factor Model, Conscientiousness, resulting in a set of 30 operators. Finally, we discuss the way to design a specific character through the use of an activation matrix, providing values for each scheme and operator.

François Bouchet, Jason M. Harley, Gregory J. Trevors & Roger Azevedo (2013). Clustering and Profiling Students According to their Interactions with an Intelligent Tutoring System Fostering Self-Regulated Learning. Journal of Educational Data Mining, 5(1), pp. 104–146.

In this paper, we present the results obtained using a clustering algorithm (Expectation-Maximization) on data collected from 106 college students learning about the circulatory system with MetaTutor, an agent-based Intelligent Tutoring System (ITS) designed to foster self-regulated learning (SRL). The three extracted clusters were validated and analyzed using multivariate statistics (MANOVAs) in order to characterize three distinct profiles of students, displaying statistically significant differences over all 12 variables used for the clusters formation (including performance, use of note-taking and number of sub-goals attempted). We show through additional analyses that variations also exist between the clusters regarding prompts they received by the system to perform SRL processes. We conclude with a discussion of implications for designing a more adaptive ITS based on an identification of learners’ profiles.

François Bouchet & Jean-Paul Sansonnet (2013). Agents Conversationnels Psychologiques : Modélisation des réactions rationnelles et comportementales des agents assistants conversationnels. Revue d'Intelligence Artificielle, 27(6), pp. 679–708.

Plusieurs études ont récemment été menées sur l’attribution de compétences cognitives et de caractéristiques psychologiques à des agents artificiels. Cependant ces études reposent sur des approches procédurales, difficiles à analyser, et elles se focalisent sur des phénomènes particuliers au lieu de couvrir une partie significative du domaine psychologique des humains. Nous présentons ici une approche systématique de l’implémentation du principe stipulant que les traits de personnalité ont une influence potentielle et effective sur le processus de décision rationnelle d’agents cognitifs. L’apport de ce travail se situe au niveau de la couverture du domaine psychologique traité et de sa généricité par rapport aux modèles d’agents rationnels utilisés. Surtout, par sa nature déclarative, il facilite l’intelligibilité des relations associant les phénomènes psychologiques aux influences sur le cycle de délibération des agents.

2012

François Bouchet & Jean-Paul Sansonnet (2012). Traits de Personnalité Computationnels: Enrichissement de la taxonomie FFM/NEO PI-R avec des gloses WordNet liées à des adjectifs de personnalité. Technique et Science Informatiques, 31(4), pp. 423–453.

We classify a set of personality-trait adjectives within the facet list of the NEO PI-R taxonomy related to the Five Factor Model. This process is based on the lexical semantics expressed by the synset-gloss attached to the adjectives in the WordNet lexical base. In order to make the arrangement of the glosses within the positions of FFM/NEO PI-R taxonomy computationally treatable, a phase of rearrangement in terms of so-called behavioral schemes is performed. This classification is synthesized as an XML resource, freely accessible on the Web, which provides a computer based support with good coverage for the study and the computational implementation of psychological behaviors in conversational agents.

2011

Evandro Manara Miletto, Marcelo Soares Pimenta, François Bouchet, Jean-Paul Sansonnet & Damián Keller (2011). Principles for Music Creation by Novices in Networked Music Environments. Journal of New Music Research, 40(3), pp. 205–216.

Networked music environments (NMEs) allow experimental artists to explore the implications of interconnecting their computers for musical purposes. Despite an evident progress in recent years of networked music research, very little attention has been paid to a very common potential kind of user: novices in music, that is, users with little or no previous music knowledge. Indeed, the same way that principles of Rich Internet Applications like YouTube and Flickr have turned the passive user into an active producer of content, we are investigating the issues to be considered by networked music environments in order to allow effective support of musical creation and experimentation by novices. CODES—a Web-based environment designed to support cooperative ways of music creation by novices—puts these principles into practice. The goal of this paper is to present, discuss and illustrate two main principles: (1) music creation by novices should be prototypical; and (2) music creation by novices should be cooperative. These principles have emerged during CODES design and development and we think they are a good starting point for further investigation of a novice-oriented perspective of NME dimensions.

François Bouchet & Jean-Paul Sansonnet (2011). Implementing WordNet Personality Adjectives as Influences on Rational Agents. International Journal of Computer Information Systems and Industrial Management Applications, 3(1), pp. 696–705.

In this paper we present a methodology dedicated to the computational implementation of personality traits in Conversational Agents. First, a significant set of personality-traits adjectives is registered from thesaurus sources. Then the lexical semantics related to personality-traits is extracted while using the WordNet database and it is given a formal representation in terms of so-called Behavioral Schemes. Finally, we propose a framework for the implementation of those schemes as influence operators controlling the decision process and the plan/action scheduling of a rational agent.

François Bouchet & Jean-Paul Sansonnet (2011). Agents Conversationnels Psychologiques : Un cadre d'étude des comportements rationnels et psychologiques des agents assistants conversationnels. Revue d'Intelligence Artificielle, 25(5), pp. 591–623.

Afin de faciliter l’accès et l’utilisation par les personnes du grand public des applications et services en informatique qui se répandent rapidement en particulier sur l’internet, il est nécessaire de proposer de nouveaux outils d’assistance qui offrent une interaction naturelle afin d’être mieux acceptés. L’approche des agents conversationnels semble prometteuse mais les agents ne peuvent pas se contenter d’opérer un raisonnement de type rationnel sur la structure et le fonctionnement des applications assistées. Ils doivent aussi exprimer des comportements psychologiques incluant des relations sociales, des traits de personnalité, des affects. Dans la première partie de l’article, nous proposons un cadre flexible pour modéliser les relations entre les réactions rationnelles et comportementales d’un agent assistant. Ensuite ce cadre est utilisé pour implémenter une première étude de cas, fondée sur la notion de biais cognitif.

2010

Mao Xuetao, Jean-Paul Sansonnet & François Bouchet (2010). Définition d’un agent conversationnel assistant d’applications Internet à partir d’un corpus de requêtes. Technique et Science Informatiques, 29(10), pp. 1123–1154.

Les agents conversationnels assistants sont une sous-classe des agents conversationnels animés, dédiée à la fonction d’assistance pour les applications et services du grand public. Les nouvelles applications internet sont un domaine particulièrement intéressant pour étudier les agents assistants pour le grand public. Nous avons donc développé un logiciel orienté web, appelé le « toolkit DIVA », où la fonction d’assistance est une question clé et où la langue naturelle joue un rôle essentiel. C’est la raison pour laquelle le toolkit DIVA repose sur une chaîne de traitement automatique de la langue naturelle (TALN) qui est chargée de traiter des requêtes d’assistance. Dans ce contexte, les outils de TALN ainsi que les outils d’assistance devraient être simples et faciles à déployer pour chaque nouvelle application web assistée par un agent DIVA. Notre proposition repose sur le recueil d’un corpus de requêtes d’assistance qui permet d’une part, de circonscrire le domaine de langue concerné et, d’autre part, d’éliciter les principaux phénomènes linguistiques qui occurrent effectivement. Sur cette base, nous avons défini une architecture de chaîne de traitement qui a été implémentée dans le toolkit DIVA ; elle a ensuite été mise à l’épreuve sur plusieurs applications test.

Conference & workshop papers

2025

Léo Nebel, François Bouchet, Vanda Luengo & Mathilde Couraud (2025). Towards Automated Characterization of Revision Events in Student Writing. Two Decades of TEL: from Lessons Learnt to Challenges Ahead.

Writing is an iterative process, yet traditional assessments often prioritize the final product over the transformations that shape it. This study investigates the complex nature of revision in student writing through computational modeling, leveraging keystroke data to capture and analyze revision behaviors. Focusing on a dataset of 1,975 annotated revision events from 10th-grade French students, we assess multiple methodological approaches, including rule-based heuristics, machine learning classifiers, and large language models (LLMs). While previous research has demonstrated the feasibility of automated revision detection, we extend these efforts by introducing a novel framework for identifying embedded revisions, i.e. instances where a revision occurs within another. By comparing the efficacy of different computational strategies, our findings reveal key insights into how revisions unfold in real-time writing. Annotations evaluated by agreement measures underline the complexity of the task. This work not only enhances the precision of automated revision classification but also lays the groundwork for intelligent writing support systems that provide targeted feedback to students, fostering a deeper engagement with the revision process.

Writing Keystroke Logging Revision

Léo Nebel, François Bouchet & Vanda Luengo (2025). Characterizing revision events in students' writing processes using LLMs. Workshop - Writing And Literacy Instruction for Educational Data Mining.

Revision is a key part of the writing process. Some important studies proposed taxonomies of revision and analysed different corpora through the lens of these taxonomies. These analyses have often been made through manual annotation after collecting the data, even if, more recently, some classifiers were trained to do this task. Based on an annotated public dataset available (ArgRewrite V.2), we go further by exploring an LLM-based classifier of revision events, which paves the way for an automatic online feedback system on the revision process students follow when writing. We compare our results to those obtained from the trained classifier of this dataset, as well as another LLM-based classifier applied in another context and discuss them. We made our code publicly available on a GitHub repository for replication.

Writing Keystroke Logging Revision

2024

Sébastien Lallé, François Bouchet, Mélina Verger & Vanda Luengo (2024). Fairness of MOOC Completion Predictions Across Demographics and Contextual Variables. International Conference on Artificial Intelligence in Education.

While machine learning (ML) has been extensively used in Massive Open Online Courses (MOOCs) to predict whether learners are at risk of dropping-out or failing, very few work has investigated the bias or possible unfairness of the predictions generated by these models. This is however important, because MOOCs typically engage very diverse audiences worldwide, and it is unsure whether the existing ML models will generate fair predictions to all learners. In this paper, we explore the fairness of ML models meant to predict course completion in a MOOC mostly offered in Europe an Africa. To do so, we leverage and compare ABROCA and MADD, two fairness metrics that have been proposed specifically in education. Our results show that some ML models are more likely to generate unfair predictions than others. Even in the fairest models, we found biases in their predictions related to how the learners’ enrolled as well as their country, gender, age and job status. These biases are particularly detrimental to African learners, which is a key finding as they are an understudied population in AI fairness analysis in education.

Demographics Fairness Machine Learning MOOCs

Valdemar Švábenský, François Bouchet, Francine Tarrazona, Michael Lopez II & Ryan S. Baker (2024). Data Set Size Analysis for Detecting the Urgency of Discussion Forum Posts. 14th International Conference on Learning Analytics & Knowledge (LAK24).

In both Massive Open Online Courses (MOOCs) and private courses, instructors face a large amount of queries in discussion forum posts that may merit a response. There has been ongoing research on how to employ machine learning to predict a post’s urgency in order to focus instructors’ attention. However, it is unclear how large a course is needed to develop these models. We took a publicly available data set of 3,503 labeled forum posts and code from one such prior study. We re-trained the six models described in the study, but with progressively smaller sample sizes, to determine if the models’ performance would be preserved. Likewise, we demonstrate that using random subsets even as small as 10% of the original data set achieves comparable performance to full data sets in five out of six models.

2023

Katia Oliver-Quelennec, François Bouchet, Thibault Carron & Claire Pinçon (2023). Usages dans le temps d'un tableau de bord d'apprentissage dans un jeu sérieux. EIAH2023 : 11ème Conférence sur les Environnements Informatiques pour l'Apprentissage Humain.

Les tableaux de bord d’apprentissage (TBA) sont des outils pour l’enseignement aux nombreux atouts. Notre étude s’intéresse à leur adaptation dans le temps, dans le cadre d’un jeu sérieux sur la pratique officinale. L’approche générique suivie consiste à (1) observer l’usage d’un TBA co-conçu et (2) questionner les étudiants sur celui-ci et sur leurs attentes dans le temps. A partir de ce jeu de données (N = 77 réponses et N = 121 traces), nous analysons les évolutions des attentes et usages selon le planning pédagogique, et confrontons ces sources d’information entre elles. Le TBA proposé a été évalué positivement par la majorité des répondants. De légères variations dans les attentes des étudiants et leurs consultations des différentes pages du TBA selon le moment du semestre apparaissent, plaidant en faveur d’une adaptation dans le temps. Ce travail montre l’importance de croiser traces d’apprentissage et avis des utilisateurs pour identifier leurs besoins, et ouvrent des perspectives de modèles d’adaptation automatiques.

Adaptation Jeu sérieux Tableau de bord d'apprentissage

Mélina Verger, Sébastien Lallé, François Bouchet & Vanda Luengo (2023). Is Your Model ”MADD”? A Novel Metric to Evaluate Algorithmic Fairness for Predictive Student Models. 16th International Conference on Educational Data Mining (EDM 2023).

Predictive student models are increasingly used in learning environments due to their ability to enhance educational outcomes and support stakeholders in making informed decisions. However, predictive models can be biased and produce unfair outcomes, leading to potential discrimination against some students and possible harmful long-term implications. This has prompted research on fairness metrics meant to capture and quantify such biases. Nonetheless, so far, existing fairness metrics used in education are predictive performance-oriented, focusing on assessing biased outcomes across groups of students, without considering the behaviors of the models nor the severity of the biases in the outcomes. Therefore, we propose a novel metric, the Model Absolute Density Distance (MADD), to analyze models' discriminatory behaviors independently from their predictive performance. We also provide a complementary visualizationbased analysis to enable fine-grained human assessment of how the models discriminate between groups of students. We evaluate our approach on the common task of predicting student success in online courses, using several common predictive classification models on an open educational dataset. We also compare our metric to the only predictive performance-oriented fairness metric developed in education, ABROCA. Results on this dataset show that: (1) fair predictive performance does not guarantee fair models' behaviors and thus fair outcomes, (2) there is no direct relationship between data bias and predictive performance bias nor discriminatory behaviors bias, and (3) trained on the same data, models exhibit different discriminatory behaviors, according to different sensitive features too. We thus recommend using the MADD on models that show satisfying predictive performance, to gain a finer-grained understanding on how they behave and regarding who and to refine models selection and their usage. Altogether, this work contributes to advancing the research on fair student models in education. Source code and data are in open access at https://github.com/melinaverger/MADD.

classification fairness metric models' behaviors sensitive features student modeling

Mélina Verger, François Bouchet, Sébastien Lallé & Vanda Luengo (2023). Caractérisation et mesure des discriminations algorithmiques dans la prédiction de la réussite à des cours en ligne. EIAH2023 : 11ème Conférence sur les Environnements Informatiques pour l'Apprentissage Humain.

Preditive models used in intelligent learning environments can suffer from biased and unfair representation. However, existing fair- ness metrics that are meant to capture these issues are only based on the models’ predictive performances. In this paper, we propose a novel fair- ness metric that measures to what extent the models behave unfairly. In addition, we provide a visualization-based analysis to qualify the types of unfair behaviors that are exhibited by the models. We apply our method on the success prediction task in online courses, with an open educational dataset. Our results highlight the need to systematically analyze unfair behaviors from the models in order to confirm or refute the sensitive nature of some attributes.

algorithmic fairness attributs sensibles équité algorithmique metric métrique sensitive attributes

Mélina Verger, Chunyang Fan, Sébastien Lallé, François Bouchet & Vanda Luengo (2023). A Fair Post-Processing Method based on the MADD Metric for Predictive Student Models. 1st International Tutorial and Workshop on Responsible Knowledge Discovery in Education (RKDE 2023) at ECML PKDD 2023.

Predictive student models are increasingly used in learning environments. However, due to the rising social impact of their usage, it is now all the more important for these models to be both sufficiently accurate and fair in their predictions. To evaluate algorithmic fairness, a new metric has been developed in education, namely the Model Absolute Density Distance (MADD). This metric enables us to measure how different a predictive model behaves regarding two groups of students, in order to quantify its algorithmic unfairness. In this paper, we thus develop a post-processing method based on this metric, that aims at improving the fairness while preserving the accuracy of relevant predictive models' results. We experiment with our approach on the task of predicting student success in an online course, using both simulated and real-world educational data, and obtain successful results. Our source code and data are in open access at https://github.com/melinaverger/MADD.

Algorithmic fairness Mitigation Success prediction

2022

Katia Oliver-Quelennec, François Bouchet, Thibault Carron & Claire Pinçon (2022). Understanding online collaboration through speech acts associated to Belbin profiles. Proceedings of the 15th International Conference on Computer-Supported Collaborative Learning - CSCL 2022.

During co-design sessions with students, we explored the link between Belbin’s profiles, roles used in project management and speech acts used to specify the speaker's intentions, which are complementary keys of successful collaboration. Identifying links between each Belbin's profile and some particular speech acts would allow us to consider automatically identifying team members' roles or recommending ways to better communicate. For this study, discourses of 14 groups have been coded in speech acts and analyzed to find links with the primary and secondary Belbin’s profiles of 31 students. We were interested in the proportion of speech acts per Belbin’s profile. We found some links as the coordinator profile intervenes less often, but answers more questions from the team or the completer-finisher profile often validates the interventions of the others members. Extending this work would involve analyzing speech acts patterns links with Belbin’s roles and how this can support successful collaborations.

Khalifa Sylla, Guy Merlin Mbatchou Nkwetchoua & François Bouchet (2022). How does the Use of Open Digital Spaces Impact Students Success and Dropout in a Virtual University?. Proceedings of the 19th International Conference on Cognition and Exploratory Learning in the Digital Age.

Virtual universities have developed considerably over the past decade, particularly on the African continent. They provide a way to deal with the considerable need to educate a large young population, but the lack of physical space can be a drawback that prevents students from succeeding and increasing dropout compared to a more traditional face-to-face university. To limit these issues, some virtual universities have been opening Open Digital Spaces (ODS) to complement the virtual space and offer students a place where to work and solve pedagogical, technical or administrative issues. However, it is unclear how students actually make use of these ODS and which uses can be beneficial or detrimental to their success and limit dropout. In this paper we lead an exploratory study of the results of a large-scale digital survey in a major African Virtual University (N=2392 answers) to identify factors in the use of Open Digital Spaces (ODS) that have an impact on students’ success and dropout. We analyzed the data using multiple Chi-Square tests of independence. Results indicate that students who visit ODSs more during the 2 weeks before an exam or only when it is mandatory are statistically less likely to succeed, contrary to students who come to work in groups or for the internet access who are more likely to succeed. Conversely, students who do not see the value of ODSs for learning and who visit only when mandatory are more likely to dropout, contrary to students who come when they have a pedagogical need or to work in groups who are less likely to dropout. Some factors particularly impact first year university students, highlighting the need to make them understand which use of the ODS are relevant to increase their chances to graduate.

Katia Oliver-Quelennec, François Bouchet, Thibault Carron, Kathy Fronton Casalino & Claire Pinçon (2022). Adapting Learning Analytics Dashboards by and for University Students. Educating for a New Future: Making Sense of Technology-Enhanced Learning Adoption.

Learning Analytics Dashboards (LADs) are becoming a key element in enabling learners to monitor their learning, plan and actually learn. However, LADs are sometimes not completely adapted to students, who are rarely involved in their design. Moreover, even when they are, the implemented LADs are often the same for all students, whereas previous works have shown the value of adapted LADs. Here we investigate which adaptations are requested by students, and attempt to identify which data and visualizations are suitable depending on the student’s profile. More specifically, we consider dynamic profiles as students’ expectations can vary over the course duration. By using LADs co-design sessions both online and on-site, we collected needs from N = 386 university students from different disciplines and degree level, split in 108 groups (2 to 4 students). After a manual annotation, we identified a total of 54 types of data and indicators, divided into 12 thematics. Our first analysis confirmed some previous results, particularly on the use of peer comparisons that do not fulfill every student’s needs. And we noticed other expectations according to the student’s learning context or the academic period. Future work will benefit from these results to define a model of adapted LADs.

Dashboard Learning analytics dashboard Co-design Indicator

2021

Thomas Sergent, François Bouchet, Morgane Daniel & Thibault Carron (2021). Using Prompts and Remediation to Improve Primary School Students Self-evaluation and Self-efficacy in a Literacy Web Application. Technology-Enhanced Learning for a Free, Safe, and Sustainable World.

Self-regulation skills are critical for students of all ages in order to maximize their learning. A key aspect of self-regulation is being aware of one’s performance and deficits in self-evaluation. Additionally, a clear consensus has not been reached regarding the age one can start learning these self-regulation processes. In order to investigate the possibility to raise awareness to some self-regulation deficits in 5 to 8 years old children, we have introduced two prompts triggered randomly after 1 out of 15 exercises into a literacy web-application for primary school students, to evaluate perceived difficulty [Too easy, Good, Too difficult] and desired difficulty [easier, same level, harder]. Comparing students’ actual performance with their responses to self-regulatory prompts can provide information about their ability to self-regulate their learning, in particular in terms of self-evaluation and self-efficacy. We collected 2,600,142 responses from 467,116 students for our experiments. The goal of this paper is to assess the impact of two different remediation strategies to reduce the two types of deficits initially measured in students.In a first study, we measured the impact of a gauge (resp. an audio recording) showing (resp. telling) the number of correct and incorrect answers to help students evaluate their actual performance during answers to the self-regulation prompts. In a second study, we measured the impact of giving self-evaluation and self-efficacy remediation to students who showed a deficit in self-regulated learning abilities from their answers to the self-regulation prompts.The results show (a) a significant reduction of self-evaluation deficits when answers were supported by a visual gauge, (b) no significant impact on self-evaluation deficits when answers were supported by an audio recording, (c) a significant reduction of future self-evaluation deficits when giving students audio feedback advising them not to repeat a detected deficit.This underlines the possibility of scaffolding self-regulated learning skills in a web based application from a young age while learning another skill.

Primary school Remediation Scaffolding Self-efficacy Self-evaluation Self-regulated learning Web based application

Camila Morais Canellas, François Bouchet, Thibaut Arribe & Vanda Luengo (2021). Towards Learning Analytics Metamodels in a Context of Publishing Chains. Proceedings of the 13th International Conference on Computer Supported Education.

In a context of pedagogical resource production via publishing chains that are based on an model-driven engineering approach, we consider the proposition of a learning analytics implementation. We argue that, by using the same approach to carry out such an implementation versus a classical one, a series of benefits could be assessed, whether they are related to the fact that it is using this specific context, methodological approach or both. Perhaps one of the most particular benefits is the detailed knowledge of the semantics and structure of any document produced, that could therefore be automatically added to the traces/analysis. Other potential improvements discussed are: separation of content and form, interoperability, compliance with data privacy, maintainability, performance, multi format, customization and reproducibility.

Camila Morais Canellas, François Bouchet & Vanda Luengo (2021). Représentations sociales de l'analytique des apprentissages avec le numérique. 10e Conférence sur les Environnements Informatiques pour l’Apprentissage Humain.

Cette étude vise à identifier les représentations sociales de l’analytique des apprentissages avec le numérique chez différents acteurs de l’éducation. Nous avons analysé les réponses de 286 participants à un questionnaire en utilisant la technique d’Évocation Hiérarchique et les résultats montrent que pour la majorité d’entre eux, les aspects les plus importants pour déterminer le domaine sont : apprentissage, données, analyse, suivi et traces. De plus, nous avons observé des variations entre différents sous-groupes de parties prenantes.

Thomas Sergent, Morgane Daniel, François Bouchet & Thibault Carron (2021). Predicting Young Students' Self-Evaluation Deficits Through Their Activity Traces. Proc. of the 14th International Conference on Educational Data Mining.

Self-evaluation is a key self-regulatory process that can already be mastered by young children. In order to assess self-evaluation skills of children, we introduced a random prompt asked randomly after 1 out of 15 exercises into a literacy web-application for primary school student, in order to evaluate the perceived difficulty [Too easy, Good, Too difficult] of the exercise they just solved. Comparing students' actual performance with their responses to this prompt can provide information about their ability to self-evaluate, and thus detect students who could improve their self-evaluation skills. We collected more than 1,000,000 responses from 300,000 students and used these data as well as performance data on each question of each exercise to predict a student's response to the next prompt, thereby estimating how likely they are to having a self-evaluation deficit. The results show (a) that a student's past responses to self-evaluation statements impacts the quality of future predictions (b) that the impact of past responses - vs their current performance - is greater when the student has low capacity for self-evaluation (c) that including older student data (answers from several sessions ago) helps in improving the accuracy of the prediction. These results pave the way (1) for adaptive polling by identifying when the model is unreliable, giving them the statement then instead of randomly, (2) for adaptive feedback, by knowing the students the most likely to show a deficit, to provide remediation.

François Bouchet & Didier Roy (2021). L’apport combiné de deux algorithmes d’IA à l’optimisation des parcours d’apprentissage dans le projet Adaptiv'Math. Dispositifs et collectifs pour la formation, l’enseignement et l’apprentissage des mathématiques.

Les algorithmes employant des techniques d’intelligence artificielle se retrouvent de plus en plus dans des solutions à destination des élèves et enseignants, mais il s’agit toutefois généralement de briques isolées. L’objectif de ce travail est de montrer comment deux algorithmes peuvent avoir des apports complémentaires dans le cadre d’une solution d’apprentissage des mathématiques en cycle 2. D’une part, un algorithme utilisant l’apprentissage par renforcement (ZPDES) personnalise le parcours de chaque élève, en suivant en temps réel sa progression et en lui proposant les exercices qui lui sont les plus adaptés. D’autre part, un algorithme de regroupement de profils (SACCOM) définit des groupes homogènes d’élèves autour de critères donnés (réussite à un exercice, nature des erreurs, …), facilitant la mise en place de pratiques pédagogiques différenciées et compensant la divergence des activités des élèves avec les parcours personnalisés. Le projet Adaptiv’Math combine ces deux approches, proposant ainsi des phases d’apprentissage en autonomie et des interventions pédagogiques de l’enseignant en petits groupes

Katia Oliver-Quelennec, François Bouchet, Thibault Carron & Claire Pinçon (2021). Faciliter le recueil de traces par la numérisation d’un outil tangible de co-design : application à la conception de tableaux de bord d’apprentissage. 10e Conférence sur les Environnements Informatiques pour l’Apprentissage Humain.

Le co-design est une approche de recueil de besoins qui s’ap- puie souvent sur des outils sous forme de cartes, mais ce format tangible, limite le recueil des données et leur exploitation. Cet article présente les facteurs clés et points de vigilance identifiés dans le cadre d’une adap- tation d’une méthode de co-design basée sur des cartes en une version numérique, à travers l’exemple de la méthode PADDLE (PArticipative Design of Dashboard for Learning in Education) pour concevoir des ta- bleaux de bord d’apprentissage (TBA). Cette adaptation numérique et l’outil associé, appelé ePADDLE, ont été testés auprès d’étudiants de première année d’université (18 groupes, N = 58). Les participants ont évalué ePADDLE comme un peu moins adapté que l’original mais tout en restant positifs. Enfin, chaque groupe a réussi à concevoir un TBA, démontrant la validité du protocole expérimental d’ePADDLE.

Thomas Sergent, Morgane Daniel, François Bouchet & Thibault Carron (2021). Détection de déficits d'auto-évaluation et d'auto-efficacité dans un logiciel enseignant la lecture et l'écriture. 10e Conférence sur les Environnements Informatiques pour l’Apprentissage Humain.

Plusieurs travaux montrent que la capacité à auto-réguler son apprentissage a un impact significatif sur les résultats scolaires. Nous présentons ici une étude visant à détecter les déficits d'auto-régulation de l'apprentissage liés à l'auto-évaluation et à l'auto-efficacité pour de jeunes (5-7 ans) élèves, dans le contexte d'une application web d'apprentissage de la lecture. Nous avons recueilli les réponses de plus de 15 000 enfants travaillant sur une telle application en classe. À partir de ces réponses, nous proposons une définition opérationnelle de différentes formes de déficits dont nous évaluons la prévalence auprès des élèves.

Katia Oliver-Quelennec, François Bouchet, Thibault Carron & Claire Pinçon (2021). Can a Learning Analytics Dashboard Participative Design Approach Be Transposed to an Online-Only Context?. Proceedings of the 18th International Conference Cognition and Exploratory Learning in Digital Age.

In-person sessions of participative design are commonly used in the field of Learning Analytics, but to reach students not always available on-site (e.g. during a pandemic), they have to be adapted to online-only context. Card-based tools are a common co-design method to collect users' needs, but this tangible format limits data collection and usage. We propose here two steps: first to use an existing co-design card deck-based method for our university context and next to adapt this new method called PADDLE (PArticipative Design of Dashboard for Learning in Education) for an online use. This article presents key factors and points of attention identified in adapting a card-based co-design method into a digital version for designing learning dashboards. This digital adaptation and the associated tool, ePADDLE, were tested with first year university students divided into 18 groups (N = 58). All groups have successfully designed a dashboard, and using the original evaluation scales, users have evaluated ePADDLE as almost as suitable as the original method. Thanks to the traces provided by the online version, we rely on speech acts to identify favorable conditions for successful collaboration.

Katia Oliver-Quelennec, François Bouchet, Thibault Carron & Claire Pinçon (2021). Analyzing the impact of e-Caducée, a serious game in pharmacy on students' professional skills over multiple years. Proc. of the 13th International Conference on Computer Supported Education.

In an academic program of our faculty of pharmacy, we tried to improve the training of future pharmacists aiming at their professionalization. We proposed a learning game called e-Caducée, which allows students to train during 3 semesters with about one hundred clinical cases. We investigated the consistency between skills worked in the game with those defined by the pedagogical team as well as the impact of the game and of the embedded dashboard on students' skills. We collected data from the game (activity traces), from the faculty (academic results) and from the students (opinion about the game). To answer our research questions, we used both multiple linear regressions as well as classical statistical inference. Results reveal that the score predictions based on the use of e-Caducée correspond with the definition of the teachers. We also found clues that the use of e-Caducée helped with learning some professional skills but the result was not confirmed with statistical analysis. Finally, we found a link between the use of the dashboard in the game and one particular professional skill’s academic results (prescription). Our future work aims at developing an adaptive learning dashboard for the game and analyzing its possible impact.

Thomas Sergent, Morgane Daniel, François Bouchet & Thibault Carron (2021). Analyse de remédiations proposées par des enseignants ciblant des déficits d'auto-régulation de jeunes élèves dans une application d'apprentissage de la lecture. 10e Conférence sur les Environnements Informatiques pour l’Apprentissage Humain.

La capacité à auto-réguler son apprentissage a un impact significatif sur les résultats scolaires. Nous avons interrogé 298 enseignants dont les élèves travaillaient avec une application web d'apprentissage de la lecture pour savoir comment ils aimeraient être informés d'éventuels déficits d'auto-régulation de leurs élèves et comment ils résoudraient ces déficits afin que la remédiation des déficits dans le logiciel d'apprentissage en ligne puisse être conçue en collaboration avec eux. Les résultats montrent un fort intérêt des enseignants pour (1) remonter les déficits d'autorégulation des élèves par le biais du tableau de bord de l'application, et (2) fournir une assistance automatisée via des encouragements et des retours formatifs, ainsi que des exercices adaptés en fonction des déficits.

Fatima Harrak & François Bouchet (2021). Aide au suivi de la progression de groupes d'apprenants pour la mise en place d'une pédagogie différenciée. 10e Conférence sur les Environnements Informatiques pour l’Apprentissage Humain.

Ce travail se situe dans le cadre d’un assistant pédagogique à destination des professeurs des écoles pour accompagner leurs élèves de primaire en mathématiques. Pour proposer à l’enseignant une vision Synthétique des activités de sa classe, nous utilisons des méthodes de clustering pour constituer automatiquement des groupes d'élèves de sa classe en fonction des activités réalisées dans l’application, où chaque élève suit un parcours individuel. Après une présentation des traces utilisées et de la qualité intrinsèque des clusters produits (coefficient de silhouette), nous analysons les retours d’une enquête auprès de 32 enseignants utilisateurs pendant au moins 6 semaines. Ces analyses ouvrent des perspectives quant aux améliorations possibles des clusters et de leur visualisation.

Thomas Sergent, Morgane Daniel, François Bouchet & Thibault Carron (2021). Addressing Children’s Self-Evaluation and Self-Efficacy Deficits in a Literacy Application. IEEE 21st International Conference on Advanced Learning Technologies.

The ability to self-regulate one’s learning (SRL) is considered to have a significant impact on educational outcomes. We present here a research work aiming first at detecting self-evaluation and self-efficacy deficits for young (5-7 years old) students, in the context of a literacy web application. From SRL answers we were able to characterize some answer patterns associated to SRL deficits, showing that around 30% of students seem to suffer from at least one of the four deficits considered in this study. We also surveyed close to 300 teachers to find out how they would like to be informed about their students’ SRL deficits and how they would address them, so that the remediation of deficits in the application could be co-designed with them.

2020

Thomas Sergent, François Bouchet & Thibault Carron (2020). Towards Temporality-Sensitive Recurrent Neural Networks through Enriched Traces. Proceedings of the 13th International Conference on Educational Data Mining.

Educational traces are distinctive compared to the usual data a recurrent neural network encounters: there is a difference between two consecutive educational traces generated by a same learner if they are separated by 2 minutes or 2 months. Indeed, in the latter case, the learner who generated the trace may have forgotten the associated skill, which is less likely in the former case. Recurrent Neural Networks have seen a surge of popularity in the recent few years thanks to Deep Knowledge Tracing. While the focus has mostly been on the network architecture, we propose here a novel framework where traces are enriched with information relative to the temporality before they are used to train the network, and assess the performance on two datasets (Lalilo and ASSISTments 2012), which is not improved by this approach.

Yves Noël, Roland Mergoil, Vanda Luengo & François Bouchet (2020). Towards a modular and flexible Learning Analytics framework. Companion Proceedings of the 10th International Conference on Learning Analytics & Knowledge LAK20.

This paper introduces a Learning Analytics platform which aims at being modular, evolving and flexible. The general framework architecture is completely independent from the digital systems to which it is connected. It collects learning data of various origins in data storages. Then it extracts a subset of the data which is aggregated into a data warehouse. Finally, these data are processed through various algorithms. Such a framework reinforces the control of data integrity in an experimental context and allows the students to refine the authorizations they give about their data. These data processing lead to indicators that will be used in student and teacher dashboards allowing a clear and fast access to learning information. In a second step, the platform will compute student profiles, facilitating the design of adaptive courses for each student.

Fatima Harrak, François Bouchet, Vanda Luengo & Pierre Gillois (2020). Evaluating Teachers' Perception of Students' Questions Organization. LAK '20: Proceedings of the Tenth International Conference on Learning Analytics & Knowledge.

Students' questions are essential to help teachers in assessing their understanding and adapting their pedagogy. However, in a flipped classroom context where many questions are asked online to be addressed in class, selecting questions can be difficult for teachers. To help them in this task, we present here three alternative ways of organizing questions: one based on pedagogical needs, one based on estimated students' profiles and one mixing both approaches. Results of a survey filled by 37 teachers in a flipped classroom pedagogy show no consensus over a single organization. A cluster analysis based on teachers' flipped classroom experience allowed us to distinguish two profiles, but they were not associated with any particular question organization preference. Qualitative results suggest the need for different organizations may rely more on a pedagogical philosophy and advocates for differentiated dashboards.

2019

Mathieu Vermeulen, Abir Karami, Anthony Fleury, François Bouchet, Nadine Mandran, Jannik Laval & Jean-Marc Labat (2019). APACHES: Human-Centered and Project-Based Methods in Higher Education. Transforming Learning with Meaningful Technologies.

Human-centered project-based teaching methods have proved their efficiency and popularity in the last decade. Such practice emphasizes the existence of interdisciplinary skills that students manipulate and incrementally learn to master throughout their higher education curriculum. This paper addresses some questions around the integration and evaluation of interdisciplinary skills. The first question focuses on the establishment of a skill-based approach to keep track of the students’ competencies over human-centered computing skills all along their curriculum. To this end, we discuss the advantages and disadvantages of existing approaches in the context of agile practices and interdisciplinary skills in human-centered project-based teaching methods. The second question deals with the tools that can accompany such approach and how they can affect the teaching courses, the university instructors’ habits and the motivation of the students. A semi-structured interviews were conducted with five instructors regarding these two questions. One main conclusion is the need to keep track of the students progress during the courses to help an efficient follow up. For this end, we propose to co-design a framework named APACHES.

Agile project Human centered computer sciences Learning analytic Skill based approach Traceability

Fatima Harrak, Vanda Luengo, François Bouchet & Rémi Bachelet (2019). Towards Improving Students' Forum Posts Categorization in MOOCs and Impact on Performance Prediction. Proceedings of the Sixth (2019) ACM Conference on Learning @ Scale.

Going beyond mere forum posts categorization is key to understand why some students struggle and eventually fail in MOOCs. We propose here an extension of a coding scheme and present the design of the associated automatic annotation tools to tag students' questions in their forum posts. Working of four sessions of the same MOOC, we cluster students' questions and show how the obtained clusters are consistent across all sessions and can be sometimes correlated with students' success in the MOOC. Moreover, it helps us better understand the nature of questions asked by successful vs. unsuccessful students.

clustering coding scheme discussion forum MOOC student's performance Student's question

François Bouchet & Rémi Bachelet (2019). Socializing on MOOCs: Comparing University and Self-enrolled Students. Digital Education: At the MOOC Crossroads Where the Interests of Academia and Business Converge.

MOOCs are becoming more and more integrated in the higher education landscape of learning, with many institutions now pushing their students towards MOOC as part of their curriculum. But what does it mean for other MOOC learners? Are these students socializing the same way when they have an easier possibility to interact with classmates offline? Is the fact that they do not personally choose to enroll in a MOOC also having an effect? In this paper, we compare university-enrolled students to other MOOC participants and in particular other self-enrolled students, to examine how and why they socialize on and around the MOOC. Using data from two French MOOCs in project management, we show that university-enrolled students are less attracted by forums and seem to interact less than others when the workload increases, which could lead to misleading conclusions when analyzing data. We therefore encourage MOOC researchers to be particularly mindful of this new trend when performing social network analyses.

Diego Oswaldo Camacho Vega & François Bouchet (2019). Self-Regulated Learning: Comparing Online and Classroom Courses in Cognition, Metacognition, Motivation, Emotions, Contexts, and Behavior. The 2019 Annual meeting of the American Educational Research Association.

This research aims at evaluating the use of cognition, metacognition, motivational, emotional, contextual and behavioral processes in self-regulated learning in online and traditional classroom environments for two separate experiments with two groups each. We used a questionnaire developed based on the adaptation of six existing scales, with the addition of a general section about the course itself. By contrasting the two experiments, results were consistent for online courses suggesting a higher mastery of motivation and positive emotions after taking the course, although it was in many ways similar to a traditional course. Finally, online course might have been associated with higher scores in context control than traditional course but it could depend of the course content.

Fatima Harrak, François Bouchet & Vanda Luengo (2019). Comparaison de questions posées et votées en ligne dans le cadre d'une classe inversée. Actes de la 9ème Conférence sur les Environnements Informatiques pour l'Apprentissage Humain.

Les questions des élèves sont utiles pour leur apprentissage et l'adaptation pédagogique des enseignants. Nous étudions ici les questions posées en ligne par des étudiants de première année de médecine, utilisées par les professeurs pour préparer des sessions de questions-réponses. Comme les étudiants peuvent aussi voter sur les questions posées, on peut s'interroger sur la valeur pédagogique du vote : a-t-il le même impact en termes d'apprentissage, et les étudiants votent-ils sur des questions similaires à celles qu'ils posent ? Pour répondre à ces questions, nous avons développé un schéma de codage de la nature des questions, puis conçu un annotateur automatique que nous avons appliqué à l'ensemble du corpus. La comparaison votants vs. non-votants révèle que les votants réussissent mieux, mais peu de différences apparaissent entre la nature des questions posées et celles votées pour les étudiants qui font les deux. Ce résultat confirme la valeur du vote comme alternative à la formulation de question pour les étudiants sachant déjà formuler leurs propres questions.

Fatima Harrak, François Bouchet & Vanda Luengo (2019). Categorizing students' questions using an ensemble hybrid approach. Proc. of the 12th International Conference on Educational Data Mining.

Students' questions categorization is a challenging task as the available corpora are often limited in size (particularly with languages other than English) and require a costly preliminary manual annotation to train the classifiers. Ensemble learning can help improve machine learning results by combining several models, and is particularly efficient to leverage the strengths of very different classifiers. In this paper, we investigate how combining a rule-based annotator (based on keywords identified by an expert) with various machine learning-based approaches and TF-IDF can improve the automated identification of questions asked by 1st year medicine students on an online platform, according to a coding scheme using 4 dimensions. First we evaluated the performance of several models, calculating the kappa between the prediction and the manually labelled dataset, according to each dimension. Then, using a stacking approach, we tried different combinations of them to design a predictive model with a higher performance. The results reveal that the new ensemble models can help to increase the performance for all dimensions of the dataset, in particular those for which the expert rule-based system showed the lowest performance. These results are promising as they indicate that some easy-to-train models can complement more manual approaches, even with a small training set of a few hundreds of annotated questions.

Fatima Harrak, François Bouchet, Vanda Luengo & Rémi Bachelet (2019). Automatic Identification of Questions in MOOC Forums and Association with Self-Regulated Learning. Proc. of the 12th International Conference on Educational Data Mining.

Discussion forums can be a rich source to analyze students' questions but it can be challenging to find relevant categories of questions. We considered here students' posts from the discussion forum of four editions of a same French MOOC on Project Management. We extended a coding scheme to annotate questions based on their content (course vs. non course) and trained 3 stages of an automatic annotation model. Then we studied the correlation between the nature of the questions asked and students' performance and self-regulation. The results are promising and reveal, for the minority of students active on forums, the possibility to use this feature to better estimate their performance and some of their self-regulation skills based on questions they ask.

MOOC coding scheme discussion forum self- regulation student's performance Student's question

2018

Guy Merlin Mbatchou, François Bouchet, Thibault Carron & Philippe Pernelle (2018). Proposing and evaluating a model of co-construction of the learning scenario by the learner. Proceedings of the 15th International Conference Cognition and Exploratory Learning in Digital Age.

To improve the learning process, the evolution of learner's characteristics (cognitive, affective, prior knowledge, workflow, organization, …) must be taken into account during the personalization or adaptation. This requires generating several scenarios (a description of activities, their order and links in the learning sequence as well as the expected outcome for the learner) adapted to the identified profiles. We propose a model which aims at improving learners' learning processes by giving them control over two key aspects: (1) the steps of the learning scenario to be followed: after each learning goal is completed, the learner chooses the next one among the possible ones (in terms of their current knowledge) while respecting pedagogical constraints (time and quality of the solutions produced according to satisfaction thresholds); (2) the assessment mode: the learner chooses a mode corresponding to their own goals in terms of mastery, while respecting the minimum thresholds set by the teacher. We assess our approach with learners in terms of (a) adequacy of the model with learners' expectations, (b) usability of the system and (c) learning experience satisfaction, through self-report questionnaires and an analysis of the data collected over 11 learners who used an implementation of our system on the LMS (Learning Management System) in the context of a real course on Economy. The results reveal an a priori acceptance of our model, a diversity of the scenarios constructed, and the use of 2 (out of 3) assessment modes to progress. We use these results to analyze current limits of the system and propose redesign ideas to minimize them.

Fatima Harrak, François Bouchet, Vanda Luengo & Pierre Gillois (2018). Profiling Students from Their Questions in a Blended Learning Environment. LAK'18: International Conference on Learning Analytics and Knowledge.

Automatic analysis of learners’ questions can be used to improve their level and help teachers in addressing them. We investigated questions (N=6457) asked before the class by 1st year medicine/pharmacy students on an online platform, used by professors to prepare their on-site Q&A session. Our long-term objectives are to help professors in categorizing those questions, and to provide students with feedback on the quality of their questions. To do so, first we manually categorized students’ questions, which led to a taxonomy then used for an automatic annotation of the whole corpus. We identified students’ characteristics from the typology of questions they asked using K-Means algorithm over four courses. The students were clustered by the proportion of each question asked in each dimension of the taxonomy. Then, we characterized the clusters by attributes not used for clustering such as the students’ grade, the attendance, the number and popularity of questions asked. Two similar clusters always appeared: a cluster (A), made of students with grades lower than average, attending less to classes, asking a low number of questions but which are popular; and a cluster (D), made of students with higher grades, high attendance, asking more questions which are less popular. This work demonstrates the validity and the usefulness of our taxonomy, and shows the relevance of this classification to identify different students’ profiles.

Guy Merlin Mbatchou, François Bouchet & Thibault Carron (2018). Multi-scenario modelling of learning. Conférence sur la Recherche en Informatique et ses Applications.

Designing an educational scenario is a sensitive and challenging activity because it is the vector of learning. However, the designed scenario may not correspond to some learners’ characteristics (pace of work, cognitive styles, emotional factors, prerequisite knowledge, …). To personalize the learning task and adapt it gradually to each learner, several scenarios are needed. Adaptation and personalization are difficult because it is necessary on the one hand to know in advance the profiles and on the other hand to produce the multiple scenarios corresponding to these profiles. Our model allows to design many scenarios without knowing the learner profiles beforehand. Furthermore, it offers each learner opportunities to choose a scenario and to change it during their learning process. The model ensures that all announced objectives have enough resources for acquiring knowledge and activities for evaluation.

e-learning adaptation instructional design learning path. learning scenario

Jason M. Harley, François Bouchet & Roger Azevedo (2018). Examining How Students’ Typical Studying Emotions Relate to Those Experienced While Studying with an ITS. Intelligent Tutoring Systems: 14th International Conference.

We help advance the research on emotions with a preliminary investigation of differences between 116 students' typical studying emotions and those they experiences while studying with an ITS. Results revealed that students reported significantly lower levels of negative emotions while studying with an ITS compared to their typical emotional dispositions toward studying.

François Bouchet, Jason M. Harley & Roger Azevedo (2018). Evaluating Adaptive Pedagogical Agents’ Prompting Strategies Effect on Students’ Emotions. Intelligent Tutoring Systems: 14th International Conference.

Adapting ITSs that promote the use of metacognitive strategies can sometimes lead to intense prompting, at least initially, to the point that there is a risk of it feeling counterproductive. In this paper, we examine the impact of different prompting strategies on self-reported agent-directed emotions in an ITS that scaffolds students’ use of self-regulated learning (SRL) strategies, taking into account students’ prior knowledge. Results indicate that more intense initial prompting can indeed lead to increased frustration, and sometimes boredom even toward pedagogical agents that are perceived as competent. When considering prior knowledge, results also show that this strategy induces a significantly different higher level of confusion in low prior knowledge students when compared to high prior knowledge students. This result is consistent with the fact that higher prior knowledge students tend to be better at self-regulating their learning, and it could also indicate that some low prior knowledge students may be on their path to a better understanding of the value of SRL.

François Bouchet, Rémi Bachelet, Hugues Labarthe & Kalina Yacef (2018). Apports d’un outil de recommandation de pairs pour lutter contre l’attrition. Actes du Colloque e-Formation 2018.

L’absence ou la perte de liens sociaux dans les MOOCs a été identifiée comme un des facteurs majeurs entraînant le décrochage d’étudiants pourtant originellement motivé pour l’obtention du certificat associé au MOOC (Yang et al., 2014). Par ailleurs, les outils promouvant la communication au sein d’un MOOC ont un impact positif sur la motivation des étudiants (Ferschke et al., 2015). Pour lutter contre ce phénomène d’attrition lié au manque de relations sociales, nous avons mis en place sur 4 sessions du MOOC GdP un outil de recommandation d’autres apprenants avec qui communiquer pour tenter de renforcer ces liens sociaux et maintenir les apprenants actifs sur la plateforme. Nous avons montré l’impact positif de cet outil en le comparant à une population n’ayant pas accès au recommandeur, même lorsque son utilisation est faible (Labarthe et al., 2016). Nous avons également comparé différentes stratégies de recommandation : (i) aléatoire ; (ii) fondée sur des critères socio-démographiques ; (iii) fondée sur la progression au sein du MOOC (Bouchet et al., 2017). Enfin nous avons étudié ce qui distingue les apprenants qui utilisent l’outil des autres, montrant qu’il permet de répondre à un besoin de communiquer d’une part des apprenants, qui n’est actuellement pas comblé malgré les autres moyens de communication déjà associés au MOOC tel que le forum ou les réseaux sociaux (Bouchet et al., 2017). L’objectif de cette communication est donc de proposer une étude comparative de l’impact des différentes versions de cet outil au cours des 4 sessions du MOOC GdP où il a été implémenté. Nous analyserons les caractéristiques des apprenants l’ayant utilisé, notamment en termes de statut socio-professionnel, capacité à l’auto-régulation et motivation, et l’effet sur le décrochage de ces différents types d’apprenants.

Hugues Labarthe, Vanda Luengo & François Bouchet (2018). Analyzing the relationships between learning analytics, educational data mining and AI for education. Proc. of learning analytics workshop at ITS 2018.

Baker and Siemens have well explained the theoretical differences and simi-larities between the educational data mining (EDM) and learning analytics (LA) communities in their 2012 seminal paper, in which they also wished for bridging the gap between both communities. Moreover, since its creation as an independent conference in 2009, EDM has been evolving in parallel with the intelligent tutoring systems (ITS) / artificial intelligence for education (AIED) community. But what are the actual links that exist between these three com-munities in terms of members and research topics: to what extent do they over-lap and work together? Are they getting closer from each other or drifting apart? Is each community specific to researchers with different backgrounds, modeling and analysis techniques? Those are some of the questions we inves-tigate using a quantitative analysis led between 2007 and 2017 through: a so-cial network analysis of the 3 communities, involving the 1822 scientists who participated in program committees and/or appeared as authors of the associat-ed journals (IJAIED, JEDM and JLA); and a text analysis of abstracts of arti-cles published in these journals. Results reveal the clear differences between these communities, their topics, practices and research methods.

Hugues Labarthe, Vanda Luengo & François Bouchet (2018). Analyse de l'hybridation entre les communautés LAK, EDM et AIED. Actes de la Journée IA pour l'Education 2018.

2017

Baptiste Monterrat, Amel Yessad, François Bouchet, Elise Lavoué & Vanda Luengo (2017). MAGAM: A Multi-Aspect Generic Adaptation Model for Learning Environments. European Conference on Technology Enhanced Learning.

Baptiste Monterrat, Amel Yessad, François Bouchet, Elise Lavoué & Vanda Luengo (2017). MAGAM: un modèle générique pour l'adaptation multi-aspects dans les EIAH. Environnements Informatiques pour l'Apprentissage Humain.

François Bouchet, Hugues Labarthe, Rémi Bachelet & Kalina Yacef (2017). Who Wants to Chat on a MOOC? Lessons from a Peer Recommender System. Digital Education: Out to the World and Back to the Campus.

Peer recommender systems (PRS) in MOOCs have been shown to help reducing attrition and increase performance of those who use them. But who are the students using them and what are their motivations? And why are some students reluctant to use them? To answer these questions, we present a study where we implemented a chat-based PRS that has been used during a MOOC session involving 6,170 students. Our analyses indicate that PRS-users are students unsatisfied by other means of interactions already available (forums, social networks…), and that they seem to use it more to share emotions than to learn together, or to assess their progression against their peers.

Baptiste Monterrat, François Bouchet, Elise Lavoué & Vanda Luengo (2017). Vers une adaptation des apprentissages générique et multi-aspects. Actes de l'atelier personnalisation et adaptation dans les environnements d'apprentissage - ORPHEE-RDV 2017.

Dans les EIAH, la personnalisation peut se faire suivant plusieurs aspects, notamment didactique, pédagogique, ludique, ou encore l’adaptation au contexte. Cet article propose un modèle d’adaptation générique dont l’apport principal est la capacité à prendre en compte de multiples aspects dans le choix d’une activité, alors que la plupart des systèmes proposés dans la littérature ne mettent en œuvre qu'un seul aspect d’adaptation. Ce modèle est basé sur l’approche de la Q-matrice. Nous souhaitons explorer certains aspects de la mise en application du modèle.

Gorgoumack Sambe, François Bouchet & Jean-Marc Labat (2017). Towards a Conceptual Framework to Scaffold Self-regulation in a MOOC. Innovation and Interdisciplinary Solutions for Underserved Areas.

MOOCs are part of the ecosystem of self-learning for which self-regulation is one of the pillars. Weakness of self-regulation skills is one of the key factors that contribute to dropout in a MOOC. We present a conceptual framework to promote self-regulated learning in a MOOC. This framework relies on the use of a virtual companion to provide metacognitive prompts and a visualization of indicators. The aim of this system will not only be to improve the quality of learning on the MOOC but also to help reducing attrition

Fatima Harrak, François Bouchet & Vanda Luengo (2017). Identifying relationships between students’ questions type and their behavior. Proc. of the 10th International Conference on Educational Data Mining.

We present the process of categorization of students’ questions, and through a clustering on students, we show the relevance of this classification to identify different profiles of students. It opens perspectives in assisting teachers during Q&A sessions.

François Bouchet, Hugues Labarthe, Kalina Yacef & Rémi Bachelet (2017). Comparing Peer Recommendation Strategies in a MOOC. ACM Extended Proceedings of UMAP 2017.

Lack of social relationship has been shown to be an important contribution factor for attrition in Massive Open Online Courses (MOOCs). Helping students to connect with other students is therefore a promising solution to alleviate this phenomenon. Following up on our previous research showing that embedding a peer recommender in a MOOC had a positive impact on stu-dents’ engagement in the MOOC, we compare in this paper the impact of three different peer recommenders.: one based on so-cio-demographic criteria, one based on current progress made in the MOOC, and the last one providing random recommenda-tions. We report our results and analysis (N= 2025 students), suggesting that the socio-demographic-based recommender had a slightly better impact than the random one.

2016

Reza Feyzi-Behnagh, Roger Azevedo, François Bouchet & Yan Tian (2016). The role of an open learner model and immediate feedback on metacognitive calibration in MetaTutor. The 2016 Annual meeting of the American Educational Research Association.

We investigated the calibration of learners’ metacognitive judgments, content page visitation, and study time allocation in the context of learning about the human circulatory system with MetaTutor, a multi-agent hypermedia learning environment. Our participants were 100 college students. Both pedagogical agent (PA) and an open learner model (OLM) were used to provide metacognitive judgment feedback to participants. The results indicated that immediate feedback led to significantly lower over- and underconfidence in participants’ metacognitive judgments and more frequent and longer duration visitations of pages relevant to their current learning sub-goal. The OLM acted as an immediate feedback mechanism for participants who didn’t receive PA’s feedback. All participants achieved significantly higher learning out comes pre-to-posttest, with no difference between with-feedback and no-feedback conditions.

Hugues Labarthe, Rémi Bachelet, François Bouchet & Kalina Yacef (2016). Increasing MOOC completion rates through social interactions: a recommendation system. Proc. of the European Stakeholder Summit on experiences and best practices in and around MOOCs.

E-learning research shows students who interact with their peers are less likely to drop out from a course, but is this applicable to MOOCs? This paper examines MOOC attrition issues and how encouraging social interactions can address them: using data from 4 sessions of the GdP MOOC, a popular Project Management MOOC, we confirm that students displaying a high level of social interaction succeed more than those who don’t. We successively explore two approaches fostering social interactions: 1) in MOOC GdP5, we gave access to private group forums, testing various group types and sizes, 2) in the ongoing MOOC GdP6, we designed a recommendation system instead, suggesting relevant chat contacts using demographic and progression criteria. We share our first results with this approach.

Hugues Labarthe, François Bouchet, Rémi Bachelet & Kalina Yacef (2016). Does a Peer Recommender Foster Students' Engagement in MOOCs?. Proc. of the 9th International Conference on Educational Data Mining.

Overall the social capital of MOOCs is under-exploited. For most students in MOOCs, autonomous learning often means learning alone. Students interested in adding a social dimension to their learning can browse discussion threads, join social medias and may decide to message other students… but usually in a random way. This common isolation might be a contributing factor on student attrition rate and on their general learning experience. To foster learners’ persistence in MOOCs, we propose to enhance the MOOC experience with a recommender which provides each student with an individual list of rich-potential contacts, created in real-time on the basis of their own profile and activities. This paper describes a controlled study conducted from Sept. to Nov. 2015 during a MOOC on Project Management. A recommender panel was integrated to the users’ interface and allowed to manage contacts, send them an instant message or consult their profile. The population (N = 8,673) was randomly split into 2 parts: a control group, without any recommendations, and an experimental group in which students could choose to activate and use the recommender. After having demonstrated that these populations were similar up to the activation of the recommender, we evaluate the effect of the recommender on the basis of four pillars of learners’ persistence: attendance, completion, success and participation. Results suggest that the recommender improved all these four factors: students were much more likely to persist and engage in the MOOC if they received recommendations than if they did not.

Niki Papaioannou, Ronald S. Landis, Cassia K. Carter, Roger Azevedo, François Bouchet & Jason M. Harley (2016). Computer-based Learning Environments in Organizational Training: Impact of Learning Environment and Personality.

The purpose of this study was to investigate the potential effects of technology and personality on training using a computer based learning environment designed to improve learning through technology-delivered instruction. The study provides a basis for considering personality characteristics when designing computer-based training interventions in organizational settings.

François Bouchet, Jason M. Harley & Roger Azevedo (2016). Can adaptive pedagogical agents’ prompting strategies improve students’ learning and self-regulation?. Intelligent Tutoring Systems: 13th International Conference.

This study examines whether an ITS that fosters the use of metacognitive strategies can benefit from variations in its prompts based on learners’ self-regulatory behaviors. We use log files and questionnaire data from 116 participants who interacted with MetaTutor, an advanced multi-agent learning environment that helps learners to develop their self-regulated learning (SRL) skills, in 3 conditions: one without adaptive prompting (NP), one with fading prompts based on learners’ deployment SRL processes (FP), and one where prompts can also increase if learners fail to deploy SRL processes adequately (FQP). Results indicated that an initially more frequent but progressively fading prompting strategy is beneficial to learners’ deployment of SRL processes once the scaffolding is faded, and has no negative impact on learners’ perception of the system’s usefulness. We also found that increasing the frequency of prompting was not sufficient to have a positive impact on the use of SRL processes, when compared to FP. These results provide insights on parameters relevant to prompting adaptation strategies to ensure transfer of metacognitive skills beyond the learning session.

2015

Jason M. Harley, Cassia K. Carter, Niki Papaioannou, François Bouchet, Ronald S. Landis, Roger Azevedo & Lana Karabachian (2015). Examining the predictive relationship between personality and emotion traits and learners’ agent-directed emotions. Proc. of the 17th Conference on Artificial Intelligence in Education (AIED 2015).

The current study examined the relationships between learners’ (N = 124) personality traits, the emotions they experience while typically studying (trait studying emotions), and the emotions they reported experiencing as a result of interacting with two Pedagogical Agents (PAs - agent-directed emotions) in MetaTutor, an advanced multi-agent learning environment. Overall, significant relationships between a subset of trait emotions (trait anger, trait anxiety) and personality traits (agreeableness, conscientiousness, and neuroticism) were found for three agent-directed emotions (pride, boredom, and neutral) though the relationships differed between the two PAs. These results demonstrate that some trait emotions and personality traits can be used to predict learners’ emotions toward specific PAs (with different roles). Suggestions are provided for adapting PAs to support learners’ (with certain characteristics) experience of positive emotions (e.g., enjoyment) and minimize their experience of negative emotions (e.g., boredom). Such an approach presents a scalable and easily implemented method for creating emotionally-adaptive, agent-based learning environments, and improving learner-PA interactions to support learning.

Nicholas Mudrick, Roger Azevedo, Michelle Taub & François Bouchet (2015). Does the Frequency of Pedagogical Agent Intervention Relate to Learners' Self-Reported Boredom while using Multi-Agent Intelligent Tutoring Systems?. Proc. of the 37th Annual Meeting of the Cognitive Science Society.

Pedagogical agents (PAs) have the ability to scaffold and regulate students' learning about complex topics while using intelligent tutoring systems (ITSs). Research on ITSs predominantly focuses on the impact that these systems have on overall learning, while the specific components of human-ITS interaction, such as student-PA dialogue within the system, are given little attention. One hundred undergraduate students interacted with MetaTutor, a multiagent hypermedia ITS, to learn about the human circulatory system. Data from these interactions were drawn from questionnaires and log-files to determine the extent to which a specific agent from MetaTutor, Sam the Strategizer, impacted students' overall emotions while using the system. Results indicated that Sam negatively impacted students' experiences of enjoyment, in relation to the other agents of MetaTutor, and the frequency of Sam's interactions with students significantly predicted their reports of boredom while using the system. Implications for the design of affect-sensitive multiagent ITSs are discussed.

François Bouchet & Rémi Bachelet (2015). Do MOOC students come back for more? Recurring Students in the GdP MOOC. Proc. of the European MOOCs Stakeholders Summit 2015.

Today, most students enrolling in a MOOC already have attended to another MOOC before. We study here a subcategory of these students, who register to the same MOOC several times, which we call Recurring Students (RS). Using data collected during three 5-week sessions of the GdP MOOC (N 14,000 on average per session), we show there is a significant and increasing proportion (over 10%) of RS. While their re-enrollment seems influenced by changes in course content, they succeed in the same proportions or less than new students (NS). A more granular analysis, separating RS who had previously completed a track (Recurring Successful Students - RSS) from those who hadn’t (Recurring Unsuccessful Students - RUS), reveals that RUS complete the new session in the same proportion as NS, whereas RSS fail more. As a conclusion, we propose recommendations to address these students, which could be valuable for other multiple-session MOOCs.

2014

Melissa Duffy, Roger Azevedo, Sarkis Meterissian & François Bouchet (2014). VirtualSelf: A computer-based learning environment for patient education using physiological and trace data.

Healthcare education systems play an important role in helping patients to understand medical information related to diagnosis, treatment options, and disease management. However, little research has explored how patients navigate biomedical information and the types of emotions elicited during learning. The purpose of this study was to examine patients’ cognitive, metacognitive, and affective processes while learning about breast cancer using VirtualSelf, a computer-based learning environment (CBLE). Diagnosed breast cancer patients (N=5) and a comparison group of healthy university students (N=5) completed several self-report measures prior to the learning session (i.e., task-value, demographics, prior knowledge, emotion regulation strategies). They then used the system for 45 minutes while we collected trace data (e.g., video of facial expressions, log-file data for navigational behaviors, and galvanic skin response for changes in arousal levels). Immediately following the session they completed several questionnaires (i.e., perceived usefulness, satisfaction, understanding). Results from product and process data focus on learning trajectories and regulatory processes across both groups. This study represents an important area of interdisciplinary research that aims to better understand how patients learn about complex disease processes. Our findings have implications for patient education, decision-making, and human factors involved in healthcare systems, including the design of CBLEs.

Roger Azevedo, Nicholas Mudrick, Michelle Taub, Reza Feyzi-Behnagh & François Bouchet (2014). Understanding the nature of self-regulated learning with metacognitive tools: Converging eye-tracking, log-file, and facial expressions. The Meeting of the EARLI Special Interest Groups 6&7 - 'Instructional Design' and 'Learning and Instruction with Computers'.

Roger Azevedo, Michelle Taub, Nicholas Mudrick, Reza Feyzi-Behnagh & François Bouchet (2014). The impact of pedagogical agents' scaffolding of metacognitive self-regulatory processes during learning with MetaTutor. The 6th Biennial Meeting of the EARLI Special Interest Group 16 - Metacognition.

Michelle Taub, Roger Azevedo, Nicholas Mudrick & François Bouchet (2014). Sub-goal sequence matters: determining the effects of sub-goal sequence on emotions during learning with hypermedia-learning environments. The 6th Biennial Meeting of the EARLI Special Interest Group 16 - Metacognition.

Nicholas Mudrick, Roger Azevedo, Michelle Taub & François Bouchet (2014). How Do Pedagogical Agents' SRL-Prompts Impact Students' Affect as They Interact with Intelligent Tutoring Systems?. The 6th Biennial Meeting of the EARLI Special Interest Group 16 - Metacognition.

Michelle Taub, Roger Azevedo, Nicholas Mudrick, Erika Clodfelter & François Bouchet (2014). Can Scaffolds from Pedagogical Agents Influence Effective Completion of Sub-Goals during Learning with a Multi-Agent Hypermedia-Learning Environment?. Learning and becoming in practice : The International Conference of the Learning Sciences (ICLS) 2014.

Jason M. Harley, François Bouchet, Niki Papaioannou, Cassia Carter, Gregory J. Trevors, Reza Feyzi-Behnagh, Roger Azevedo & Ronald S. Landis (2014). Assessing Learning with MetaTutor, a Multi-Agent Hypermedia Learning Environment. The 2014 Annual meeting of the American Educational Research Association.

In this paper we discuss the ways in which assessments of learning were evaluated, designed, and implemented in MetaTutor (a multi-agent, hypermedia learning environment about several human body systems; Azevedo et al., 2012, 2013). We also share the lessons that we have learned from assessing learning with MetaTutor across three different universities and three studies (N = 336). Techniques and considerations for assessing learning are shared, including pedagogical and psychometric properties of items. Results revealed significant differences in pre to posttest learning across all studies and that changes to the content and assessments of learning lead to lower scores, in particular, on the pretest where cores had previously been high and limited variance. The methods and results of this paper can be used to motivate other researchers who use CBLEs to examine and improve the psychometric and pedagogical features of their learning assessments.

Roger Azevedo, Nicholas Mudrick, Michelle Taub, Reza Feyzi-Behnagh & François Bouchet (2014). Are pedagogical agents effective in scaffolding metacognitive processes during learning with MetaTutor?. The 6th Biennial Meeting of the EARLI Special Interest Group 16 - Metacognition.

John Ranellucci, François Bouchet, Eric G. Poitras, Susanne P. Lajoie & Roger Azevedo (2014). An analysis of emotions in educationally relevant tweets. The 2014 Annual meeting of the American Educational Research Association.

Nicholas Mudrick, Roger Azevedo, Michelle Taub, Reza Feyzi-Behnagh & François Bouchet (2014). A Study of subjective emotions, self-regulatory processes, and learning gains: are pedagogical agents effective in fostering learning?. Learning and becoming in practice : The International Conference of the Learning Sciences (ICLS) 2014.

Though some research has focused on agent-direct affective processes, none has examined its impact on multi-agent learning environments and on the detection, modeling and fostering of self-regulated learning processes. 38 participants interacted with MetaTutor, an intelligent, multi-agent hypermedia-learning environment, to learn about the human circulatory system. The log files, containing information about their overall performance, and self-report measures, assessing emotions and impressions towards agents obtained from their interactions with MetaTutor were used to assess the relationship between subjective agent-directed emotions, SRL processes and overall learning gains. Results indicate that agent-directed emotions were not significantly related to SRL strategy use, negative agent-directed emotions were significantly related to negative learning gains and negative agent-directed emotions for two specific agents (representative of two SRL pillars) were related to negative learning gains. Implications for the design of multi-agent systems and the role of emotions during human-agent interactions and their relation to learning are discussed.

Jason M. Harley, François Bouchet, M. Sazzad Hussain, Roger Azevedo & Rafael A. Calvo (2014). A Multi-Componential Analysis of Emotions during Complex Learning with an Intelligent Multi-Agent System. The 2014 Annual meeting of the American Educational Research Association.

In this paper we discuss the methodology and results of aligning three different emotional measurement methods (automatic facial expression recognition, self-report, electrodermal activation) and their agreement regarding learners’ emotions. Data was collected from 67 undergraduate students from a North American university who interacted with MetaTutor, an intelligent, multi-agent, hypermedia environment for learning about the human circulatory system, for a 1 hour learning session (Azevedo et al., 2013, Harley, Bouchet, & Azevedo, 2013). A webcam was used to capture videos of learners’ facial expressions, which were analyzed using automatic facial recognition software (FaceReader 5.0). Learners’ physiological arousal was measured using Affectiva’s Q-Sensor 2.0 electrodermal activation bracelet. Learners self-reported their experience of 19 different emotional states (including basic, learner-centered, and academic achievement emotions) using the Emotion-Value questionnaire (Harley et al., 2013). They did so on five different occasions during the learning session, which were used as markers to align data from FaceReader and Q-Sensor. We found a high agreement between the facial and self-report data (75.6%) when similar emotions were grouped together along theoretical dimensions and definitions (e.g., anger and frustration) (Harley, et al., 2013). However, our new results examining the agreement between the Q-Sensor and these two methods suggests that electrodermal (EDA/physiological) indices of emotions do not have a tightly coupled (Gross, Sheppes, & Urry, 2011) relationship with them. Explanations for this finding are discussed.

2013

Roger Azevedo, Jason M. Harley, François Bouchet, Reza Feyzi-Behnagh, Michelle Taub, Gregory Trevors & Melissa Duffy (2013). MetaTutor: an innovative technology environment to study and assess self-regulatory processes. Symposium in the 15th Biennial Meeting of the European Association for Research on Learning and Instruction (EARLI 2013).

Babak Khosravifar, Roger Azevedo, Reza Feyzi-Behnagh, François Bouchet, Jason M. Harley, Melissa Duffy, Gregory J. Trevors & Michelle Taub (2013). Using Intelligent Multi-Agent Systems to Model and Foster Self-Regulated Learning: A Theoretically-Based Approach using Markov Decision Process. The 2013 Annual meeting of the American Educational Research Association.

Babak Khosravifar, François Bouchet, Reza Feyzi-Behnagh, Roger Azevedo & Jason M. Harley (2013). Using Intelligent Multi-Agent Systems to Model and Foster Self-Regulated Learning: A Theoretically-Based Approach using Markov Decision Process. The IEEE 27th International Conference on Advanced Information Networking and Applications (AINA-2013).

In self-regulated learning concept, Intelligent Tutoring Systems (ITS) can be designed to foster learning behaviors through pedagogical agents (PAs) that are used for interactions and exchange information with the human learner. These agents are intelligent and follow rational behaviors, but in the case of multi-agent environments they need to be systematically and specifically designed, however in order to follow a common goal, different self-regulatory systems have been designed that use pedagogical agents, but they fail to constrain the decision making of the agents and maintain a sequential decision making process during learning interactions with human learners. In this paper, we provide a new theoretical model for agent-learner interactions in MetaTutor, a multi-agent hypermedia learning environment, using Markov decision processes. We theoretically define the agents' Markov decisions and their influence on MetaTutor's performance as a whole. First, we formally define the Markov architecture and its parameters. We then link these characteristics to the pedagogical agents we use in MetaTutor and define different versions of MetaTutor agents equipped with Markov decision mechanism. Furthermore, we explore additional details about agents' sequential decision making and how reward functions influence their acting strategies with learners in the learning environment. We introduce the optimization problem in which we aim to maximize the expected return of the overall agents' acts in a self-regulatory system. What specifically distinguishes this work from the previous proposals in the same domain is its novelty in continuous decision making mechanism investigation and performance analysis that improve the applicability of the proposed adaptive model in a multi-agent ITS like MetaTutor.

Reza Feyzi-Behnagh, Gregory J. Trevors, Roger Azevedo, Wook Yang, Valérie Bélanger-Cantara, Joan Henchey, François Bouchet, Nicole Pacampara & Grace Wang (2013). Understanding Multimedia Learning by Converging Process and Product Data. The 2013 Annual meeting of the American Educational Research Association.

Integrating text and diagrams during multimedia learning is extremely challenging for students and a major issue impeding metacomprehension of scientific material. The effect of three types of discrepancies (within-text [WT], between-text-and-graph [BTG], and no-discrepancy) on participants’ multimedia learning of science content, inspection time, and the use of learning strategy “coordination of informational sources” (COIS) was investigated. Forty-two (N = 42) college students were asked to study 12 multimedia pages of science material (comprised of text and graphs) and answered inference questions based on the content. Results indicated that participants used significantly more instances of COIS strategy on pages with within-text discrepancy. They inspected graphs significantly longer on pages with BTG discrepancy. Finally, participants had significantly lower scores on inference questions for pages with a BTG discrepancy. This study has implications for research with authentic multimedia science content and for the design of effective multimedia learning.

Gregory J. Trevors, Reza Feyzi-Behnagh, Anoop Saxena, François Bouchet & Roger Azevedo (2013). Students regulate their learning processes as a function of multimedia coherence: Analyses of eye-gaze behaviour. The 15th Biennial Meeting of the European Association for Research on Learning and Instruction (EARLI 2013).

This study investigated students’ regulation of learning processes across science multimedia. Within an adaptive, hypermedia learning environment, 81 university students were presented 38 pages of texts and images of the circulatory system. A subset of 4 pages were rated for high or low coherence between text and image. Participants’ learning strategy behaviours (i.e., seeking out, attending to, and integrating relevant multimedia content) were assessed by aligning eyetracking and computer-based learning environment log-files. Subsequent learning achievement was assessed with a multiple choice posttest. Result show that students regulated their learning processes as a function of text-image coherence, which was related to learning achievement: processes on high-coherence pages positively related to learning, whereas processes on lowcoherence pages negatively related to learning. Implications for theory and educational practice are discussed.

Jason M. Harley, François Bouchet & Roger Azevedo (2013). Providing adaptive, real-time tutorial feedback in MetaTutor. Annual Conference of the Education Graduate Students' Society: Meeting in the Middle: (de)Constructing Knowledge.

MetaTutor is a hypermedia, multi-agent learning environment that teaches students about the human circulatory system and fosters the development of self-regulated learning skills. MetaTutor’s design is guided by contemporary models of SRL that emphasize the temporal deployment of cognitive, metacognitive, and affective (CAM) processes during learning and treat SRL as an event. One of the greatest challenges to developing real-time feedback protocols for intelligent tutoring systems’ pedagogical agents is using and aligning complex, multi-channel data including self-report, log file, physiological and behavioral data. We outline some of the different data channels we are using, how we are using them to innovate upon the design and programming of MetaTutor agent feedback, and highlight future directions.

Jason M. Harley, Michelle Taub, François Bouchet, Joan Henchey & Roger Azevedo (2013). Profiling Learners’ Co-Regulation Patterns with a Pedagogical Agent in an Intelligent Tutoring System for Learning About Human Biology. The 2013 Annual meeting of the American Educational Research Association.

Examining co-regulated learning between human and artificial pedagogical agents is extremely important and has recently led to a surge in research on the conceptual, theoretical, methodological, analytical, and educational issues behind other-regulated learning. This study used a novel coding framework to examine Co-Regulated Learning (CoRL) between learners and pedagogical agents (PAs) during the sub-goal setting phase with MetaTutor, an adaptive, intelligent, hypermedia learning environment. This paper highlights the importance of examining tutorial feedback using a granular research lens, which we used to examine differences in the interactions (and learners’ subsequent learning sessions) between learners and a PA both within and between MetaTutor’s two pedagogical agent scaffolding conditions: prompt and feedback (PF) and control (C). The results demonstrate that co-regulated learning interactions can take place between learners and PAs and that they can lead to increased use of SRL strategies (judgments of learning, content evaluation), a more effective use (e.g., selection of more relevant pages related to one’s goals), as well as higher proportional learning gains than less collaborative learner-PA interactions. Future work will examine the relationship between different learner-PA interactions and emotions as well as additional co-regulated learning interactions between PAs and learners.

Jean-Paul Sansonnet, François Bouchet & Nicolas Sabouret (2013). Moteurs de personnalité pour agents dialogiques : étude d’un modèle pour les traits interpersonnels. SMA Systèmes Multi-agents JFSMA 13 - 21e Journées francophones sur les systèmes multi-agents.

Nous présentons une approche flexible et déclarative pour l’implémentation informatique de traits de personnalité issus de la taxonomie FFM / NEOPI-R . Elle repose sur le principe que les traits activent des opérateurs influençant le processus de décision rationnelle des agents. Notre approche est validée sur une étude de cas portant sur la sous-classe des traits interpersonnels, appliquée aux agents dialogiques. — We present a flexible and declarative approach to the implementation of personality traits from the FFM / NEOPI-R taxonomy. This approach is based on the principle that personality traits influence the rational decision making process of the agents. The proposition is validated upon a case study involving the sub-class of interpersonal traits, applied to dialogical agents.

Jean-Paul Sansonnet & François Bouchet (2013). Managing personality influences in dialogical agents. Proceedings of the 5th International Conference on Agents and Artificial Intelligence (ICAART 2013).

We present in this article an architecture implementing personality traits from the FFM/NEO PI-R taxonomy as influence operators upon the rational decision making process of dialogical agents. The objective is to separate designer-dependent resources (traits taxonomies, influence operators, behaviors/operators links) from the core part of the computational implementation (the personality engine). Through a case study, we show how our approach makes it easier to combine various resources and to observe various scenarios within a single framework.

Daria Bondareva, Cristina Conati, Reza Feyzi-Behnagh, Jason M. Harley, Roger Azevedo & François Bouchet (2013). Inferring Learning from Gaze Data during Interaction with an Environment to Support Self-Regulated Learning. Proc. of the 16th Conference on Artificial Intelligence in Education (AIED 2013).

In this paper, we explore the potential of gaze data as a source of information to predict learning as students interact with MetaTutor, an ITS that scaffolds self-regulated learning. Using data from 47 college students, we show that a classifier using a variety of gaze features achieves considerable accuracy in predicting student learning after seeing gaze data from the complete interaction. We also show promising results on the classifier ability to detect learning in real-time during interaction.

François Bouchet, Jason M. Harley & Roger Azevedo (2013). Impact of Different Pedagogical Agents’ Adaptive Self-Regulated Prompting Strategies on Learning with MetaTutor. Proc. of the 16th Conference on Artificial Intelligence in Education (AIED 2013).

Extended interactions with a pedagogical agent (PA) assisting students to enact cognitive and metacognitive self-regulated processes requires the system to adapt the types and frequency of scaffolding. We compared learners’ perception of PAs’ prompts with MetaTutor, a hypermedia adaptive learning environment, with 40 undergraduates randomly assigned to one of three conditions: non-adaptive prompting (NP), frequency-based adaptive prompting (FP) and frequency and quality-based adaptive prompting (FQP). Results indicate learners are unable to reliably perceive differences in the number of prompts received, though these differences are reflected in positive outcomes in terms of SRL processes enacted and learning gains, and negative outcomes in terms of self-reported satisfaction. Preliminary results indicated that more frequent, but adaptive prompting is an efficient scaffolding strategy, despite negatively impacting learners’ satisfaction.

Gregory J. Trevors, Reza Feyzi-Behnagh, Roger Azevedo, Wook Yang, Joan Henchey, Valérie Bélanger-Cantara, François Bouchet, Grace Wang & Nicole Pacampara (2013). Eye-Movement Patterns in Science Multimedia as a Function of Epistemic Beliefs and Learning Task Conditions. The 2013 Annual meeting of the American Educational Research Association.

The aim of the current study was to extend research on the effects and interactions between epistemic beliefs and learning task conditions by investigating students' on-line regulation of attentional allocation evidenced in eye-movement patterns during multimedia science learning.

Jason M. Harley, Cassia Carter, Niki Papaioannou, François Bouchet, Roger Azevedo & Ronald S. Landis (2013). Examining Learners’ Academic Achievement Emotions during Science Learning with an Intelligent, Hypermedia Multi-Agent System. The 2013 Annual meeting of the American Educational Research Association.

We used Pekrun et al.’s (2002) AEQ measure to investigate the relationship between college students’ (N = 105) trait emotions and their emotional experiences with MetaTutor, an intelligent, multiagent, hypermedia environment for learning about the human circulatory system. We also examined the relationship between learners’ self-reported emotions and learning outcomes using their proportional learning gains. We also examined whether learners’ emotional experiences significantly different between MetaTutor’s two pedagogical agent (PA) scaffolding conditions: prompt and feedback (PF) and control (C). Results revealed that learners’ trait and state emotions significantly differed, but that no significant differences existed between the two pedagogical agent conditions. We also found that emotions were not generally related to learning outcomes. These findings suggest that MetaTutor may have a dampening effect on learners’ negative achievement emotions and prompts further study into the potential relationship between learning and the experience of neutral states.

Michelle Taub, Roger Azevedo, François Bouchet, Reza Feyzi-Behnagh & Jason M. Harley (2013). Can prior knowledge adequately predict the use of metacognitive processes during hypermedia learning?. The 15th Biennial Meeting of the European Association for Research on Learning and Instruction (EARLI 2013).

Roger Azevedo, Reza Feyzi-Behnagh, Jason M. Harley & François Bouchet (2013). Analyzing temporally unfolding self-regulatory processes during learning with multi-agent technologies. Symposium in the 15th Biennial Meeting of the European Association for Research on Learning and Instruction (EARLI 2013).

Reza Feyzi-Behnagh, Gregory J. Trevors, François Bouchet & Roger Azevedo (2013). Aligning Multiple Sources of SRL Data in MetaTutor: Towards Interactive Scaffolding in Multi-Agent Systems. Symposium in the 15th Biennial Meeting of the European Association for Research on Learning and Instruction (EARLI 2013).

100 undergraduate participants were randomly assigned to one of two experimental conditions (Prompt and Feedback [PF] and Control), and used MetaTutor (a multi-agent hypermedia intelligent tutoring system [ITS]) to learn about a challenging science topic (i.e., the human circulatory system) for two hours. During the session, we collected product (e.g., pretest, posttest, quizzes, summaries), and process (e.g., concurrent think-alouds, eye-tracking, log-files, face videos, physiological data, screen recordings, metacognitive judgments, and notes and drawings) data to analyze the roles of cognitive and metacognitive processes during learning about the topic with the system. Aligning trace data from eye-tracking and system-generated log-files, we investigated the occurrence of the SRL strategy coordination of informational sources (COIS), text and diagram inspection time, as well as the occurrence and accuracy of metacognitive judgments (JOLs, FOKs, and CEs). Results indicate that learners in the PF condition performed significantly higher number of COIS compared to those in the control group, and had significantly more accurate metacognitive judgments. We will illustrate and describe how these data, methodology, and findings can be used in augmenting current models of SRL, and ultimately in the design of ITSs which can provide adaptive scaffolding for interacting learning of challenging science topics, such as the human circulatory system.

Jason M. Harley, François Bouchet & Roger Azevedo (2013). Aligning and Comparing Data on Emotions Experienced during Learning with MetaTutor. Proc. of the 16th Conference on Artificial Intelligence in Education (AIED 2013).

In this study we aligned and compared self-report and on-line emotions data on 67 college students’ emotions at five different points in time over the course of their interactions with MetaTutor. Self-reported emotion data as well as facial expression data were converged and analyzed. Results across channels revealed that neutral and positively-valenced basic and learner-centered emotional states represented the majority of emotional states experienced with MetaTutor. The self-report results revealed a decline in the intensity of positively-valenced and neutral states across the learning session. The facial expression results revealed a substantial decrease in the number of learners’ with neutral facial expressions from time one to time two, but a fairly stable pattern for the remainder of the session, with participants who experienced other basic emotional states, transitioning back to a state of neutral between self-reports. Agreement between channels was 75.6%.

2012

Roger Azevedo, François Bouchet, Jason M. Harley, Reza Feyzi-Behnagh, Gregory Trevors, Melissa Duffy, Michelle Taub, Nicole Pacampara, Lauren Agnew, Sophie Griscom, Nicholas Mudrick, Victoria Stead & Wook Yang (2012). MetaTutor: An Intelligent Multi-Agent Tutoring System Designed to Detect, Track, Model, and Foster Self-Regulated Learning. Proc. of the Fourth Workshop on Self-Regulated Learning in Educational Technologies.

MetaTutor is both (1) a learning tool designed to teach and train students to self-regulate (e.g., by modeling and scaffolding metacognitive monitoring, facilitating the use of effective learning strategies, and setting and coordinating relevant learning goals), and (2) a research tool used to collect trace data on students’ cognitive, metacognitive, affective, and motivational processes de-ployed during learning. In this session we will showcase MetaTutor, an adap-tive multi-agent ITS and demonstrate its ability to detect, track, model, and fos-ter self-regulated learning (SRL). In addition, we will also provide: (1) a brief overview of the theoretical and empirical basis of the system; (2) an overview of the system’s ability to detect, track, model, and foster learners’ SRL; (3) an overview of the data types collected during learning (e.g., concurrent think-alouds, eye-tracking, note taking and drawing, log-files, and facial detection of emotions) and describe how these data are used to (4) make inferences regard-ing the system’s ability to model, scaffold, and foster learners’ SRL.

Roger Azevedo, François Bouchet, Reza Feyzi-Behnagh, Jason M. Harley, Melissa Duffy & Gregory J. Trevors (2012). MetaTutor as an innovative technology environment to assess students’ self-regulatory processes. Symposium on Knowing What Students Know and Feel: Innovative Technology Rich Assessments at the 2012 Annual meeting of the American Educational Research Association.

John Ranellucci, Eric Poitras, François Bouchet, Susanne P. Lajoie & Roger Azevedo (2012). Using social networking to guide research on emotions in education. The 2012 Annual Conference of the Canadian Society for the Study of Education.

Roger Azevedo, Jason M. Harley, Reza Feyzi-Behnagh & François Bouchet (2012). Using on-line measures to understand self-regulated learning with advanced learning technologies. Symposium on Integrating Different Approaches to Investigating Self-Regulated Learning at the 2012 Annual meeting of the American Educational Research Association.

François Bouchet & Jean-Paul Sansonnet (2012). Une approche facilitant la couverture et l’intelligibilité des influences des traits de personnalité sur le raisonnement rationnel des agents. SMA Systèmes Multi-agents JFSMA 12 - 20e Journées francophones sur les systèmes multi-agents.

Nous présentons une approche systématique de l’implémentation du principe stipulant que les traits de personnalité ont une influence potentielle et effective sur le processus de décision rationnelle d’agents cognitifs. L’apport de ce travail se situe au niveau de la couverture du domaine psychologique traité, de sa généricité par rapport aux modèles d’agents rationnels utilisés, et surtout, par sa nature déclarative, il facilite l’intelligibilité des relations associant les phénomènes psychologiques aux influences sur le cycle de délibération des agents.

Roger Azevedo, Ronald S. Landis, Reza Feyzi-Behnagh, Melissa Duffy, Gregory J. Trevors, Jason M. Harley, François Bouchet, Jonathan Burlison, Michelle Taub, Nicole Pacampara, Mohammed Yeasin, A. K. Mahbubur M. Rahman, M. Iftekhar Tanveer & Gahangir Hossain (2012). The Effectiveness of Pedagogical Agents’ Prompting and Feedback in Facilitating Co-Adapted Learning with MetaTutor. Intelligent Tutoring Systems, 11th International Conference.

Co-adapted learning involves complex, dynamically unfolding interactions between human and artificial pedagogical agents (PAs) during learning with intelligent systems. In general, these interactions lead to effective learning when (1) learners correctly monitor and regulate their cognitive and metacognitive processes in response to internal (e.g., accurate metacognitive judgments followed by the selection of effective learning strategies) and external (e.g., re-sponse to agents’ prompting and feedback) conditions, and (2) pedagogical agents can adequately and correctly detect, track, model, and foster learners’ self-regulatory processes. In this study, we tested the effectiveness of PAs’ prompting and feedback on learners’ self-regulated learning about the human circulatory system with MetaTutor, an adaptive, multi-agent learning environ-ment. Sixty-nine (N=69) undergraduates learned about the topic with MetaTutor, during a 2-hour session under one of three conditions: prompt and feedback (PF), prompt-only (PO), and no prompt (NP) condition. The PF condition re-ceived timely prompts from several pedagogical agents to deploy various SRL processes and received immediate directive feedback concerning the deployment of the processes. The PO condition received the same timely prompts, without feedback. Finally, the NP condition learned without assistance from the agents. Results indicate that those in the PF condition had significantly higher learning efficiency scores than those in both the PO and control conditions. In addition, log-file data provided evidence of the effectiveness of the PA’s timely scaffolding and feedback in facilitating learners’ (in the PF condition) metacognitive monitoring and regulation during learning.

Jason M. Harley, François Bouchet & Roger Azevedo (2012). Measuring Learners' Co-Occurring Emotional Responses during their Interaction with a Pedagogical Agent in MetaTutor. Intelligent Tutoring Systems, 11th International Conference.

This paper extends upon traditional emotional measurement frameworks used by ITSs in which emotions are analyzed as single, discrete psychological experiences by examining co-occurring emotions (COEs) (e.g., Conati) through a novel methodological approach. In this paper we examined the occurrence of students’ embodiment of basic single discrete emotions (SDEs) and COEs (in addition to neutral) using an automatic facial expression recognition program, FaceReader 4.0. This analysis focuses on the sub goal setting task of learners’ (N = 50) interaction with MetaTutor, during which a pedagogical agent assisted students to set three relevant sub goals for their learning session. Results indicated that neutral and sadness were the SDEs experienced most by students and also the most represented emotions in COE pairs. COEs represented nearly a quarter of students’ embodied emotions.

Roger Azevedo, Reza Feyzi-Behnagh, Jason M. Harley, François Bouchet, Gregory J. Trevors, Melissa Duffy & Zaynab Sabagh (2012). Measuring self-regulated learning with a multi-agent hypermedia environment. Symposium on Measuring Self-Regulated Learning with Multi-Agent Learning Environments at the 2012 Annual meeting of the American Educational Research Association.

Jason M. Harley, François Bouchet & Roger Azevedo (2012). Measuring learners’ unfolding, discrete emotional responses to different pedagogical agents scaffolding strategies. The 2012 Annual meeting of the American Educational Research Association.

This study investigates learners’ discrete emotional responses to a pedagogical agent’s (PA) tutorial scaffolding strategies during their interactions with it while setting three sub goals for their learning session with MetaTutor. 24 undergraduate students from a large, research-intensive university participated in this study and were randomly assigned to one of the two scaffolding conditions (Feedback and Non-Feedback). Results revealed that the PA scaffolding strategy exerted a significant main effect on participants’ embodiment of positive emotions (happiness) and that there was a significant interaction effect of PA scaffolding strategy and sub goal (within subjects measure) on participants’ embodiment of neutrality. These results support the design of a tutorial strategy that is able to both scaffold learning and support adaptive emotional regulation.

François Bouchet, John S. Kinnebrew, Gautam Biswas & Roger Azevedo (2012). Identifying Students' Characteristic Learning Behaviors in an Intelligent Tutoring System Fostering Self-Regulated Learning. Proceedings of the 5th International Conference on Educational Data Mining.

Identification of student learning behaviors, especially those that characterize or distinguish students, can yield important insights for the design of adaptation and feedback mechanisms in Intelligent Tutoring Systems (ITS). In this paper, we analyze trace data to identify distinguishing patterns of behavior in a study of 51 college students learning about a complex science topic with an agent-based ITS that fosters self-regulated learning (SRL). Preliminary analysis with an Expectation-Maximization clustering algorithm revealed the existence of three distinct groups of students, distinguished by their test and quiz scores (low for the first group, medium for the second group, and high for the third group), their learning gains (low, medium, high), the frequency of their note-taking (rare, frequent, rare) and note-checking (rare, rare, frequent), the proportion of sub-goals attempted (low, low, high), and the time spent reading (high, high, low). In this paper, we extend this analysis to identify characteristic learning behaviors and strategies that distinguish these three groups of students. We employ a differential sequence mining technique to identify differentially frequent activity patterns between the student groups and interpret these patterns in terms of relevant learning behaviors. The results of this analysis reveal that high-performing students tend to be better at quickly identifying the relevance of a page to their subgoal, are more methodical in their exploration of the pedagogical content, rely on system prompts to take notes and summarize, and are more strategic in their preparation for the post-test (e.g., using the end of their session to briefly review pages). These results provide a first step in identifying the group to which a student belongs during the learning session, thus making possible a real-time adaptation of the system.

Roger Azevedo, Reza Feyzi-Behnagh, Jason M. Harley, François Bouchet & Michelle Taub (2012). Externally-Regulated Learning between Human and Artificial Pedagogical Agents in the Context of a Multi-Agent Adaptive Hypermedia Environment. Symposium on Innovations in Researching Regulation of Learning in Solo and Collaborative Tasks at the 2012 Annual meeting of the American Educational Research Association.

Jason M. Harley, François Bouchet & Roger Azevedo (2012). Co-Occurring Emotions: Building an Understanding of Parallel-Emotional Processing and its Applications to Learning and Education with Intelligent Tutoring Systems. Annual Conference of the Education Graduate Students' Society: (e)Merging Knowledges: Classroom, Community, Culture.

Currently, very little is known about co-occurring (i.e., parallel) emotional states, despite several researchers accounting for them in theories and identifying them as crucial to the cross-disciplinary understanding and measurement of emotions as well as in the programming of emotionally-adaptive intelligent tutoring systems (ITS) (Conati, 2009; Ekman & Friesen, 1971; Pekrun, 2006). This paper expands upon the traditional emotional measurement and conceptual framework in which emotions are analyzed and reported as single, discrete psychological experiences by: (1) using a novel methodological trace-data approach in which co-occurring emotions are examined using an automatic facial recognition program, FaceReader 4.0; (2) measuring state-transitions between single and co-occurring emotions (e.g., “neutral” to “happy and surprised” to “scared”); and (3) examining co-occurring emotional antecedents (e.g., pedagogical agent feedback). In the paper we discuss our findings and the theoretical conclusions we can draw from them, in particular, whether there are multiple identifiable ‘types’ of co-occurring emotions (e.g., hybrid states, transition-based, or competing or complimentary goal/stimuli-directed states?). The implications these results have for the development of emotionally adaptive ITS and their future implementation as classroom and homework teaching assistants (i.e., tutors) will be discussed.

Jason M. Harley, Michelle Taub, François Bouchet & Roger Azevedo (2012). A Framework to Understand the Nature of Co-Regulated Learning in Human-Pedagogical Agent Interactions. Proc. of the Fourth Workshop on Self-Regulated Learning in Educational Technologies.

Contemporary research has examined the learning process through the lens of self-regulated learning (SRL) while relatively little effort has been given to examining human tutors’ (HTs) or pedagogical agents’ (PAs) role in supporting learners’ development of SRL strategies. This paper addresses this shortcoming by outlining the design of an emerging coding scheme that facili-tates the examination of co-regulated learning (CoRL) between a human learner and artificial PAs within the context of learning with MetaTutor, an intelligent tutoring system (ITS).

2011

Jean-Paul Sansonnet & François Bouchet (2011). Personnification d'Entités par Agents Conversationnels. Actes des Sixièmes Journées Francophones Modèles Formels de l'Interaction (MFI'11).

La question de l'accès à l'information ne se pose plus tant en termes de la production et du transport qu'en termes d'accroche, d'acceptabilité et d'assimilation auprès du grand public qui utilise massivement l'Internet. Pour ce faire, nous proposons une approche fondée sur la notion d'interaction dialogique entre un usager novice et un agent conversationnel chargé de médier des entités computationnelles. L'originalité de l'approche repose sur le principe de personnalisation d'une entité au moyen d'un agent qui s'identifie à elle.

Jean-Paul Sansonnet, François Bouchet & William Turner (2011). Personification of Topics with Conversational Agents. Proc. of Interfaces and Human-Computer Interaction 2011.

Nowadays, ordinary people can freely access a large amount of information through Internet. Then issues of production and transport of computational entities are superseded by new requirements like: enticement, acceptability and understanding. To achieve these goals, we propose an approach based on dialogical interaction between novice users and conversational agents achieving the mediation of computational entities. The originality of the approach is based on the principle of personification of an entity by an agent, where the agent and the entity share a single identity.

Jean-Paul Sansonnet & François Bouchet (2011). Integrating Psychological Behaviors in the Rational Process of Conversational Assistant Agents. Proceedings of the Twenty-Fourth International Florida Artificial Intelligence Research Society Conference (FLAIRS 2011).

In this paper, we describe a framework dedicated to studies and experimentations upon the nature of the relationships between the rational reasoning process of an artificial agent and its psychological counterpart, namely its behavioral reasoning process. This study is focused on the domain of Conversational Assistant Agents, which are software tools providing various kinds of assistance to people of the general public interacting with computer-based applications or services. In this context, we show on some examples the need for the agents to be able to exhibit both a rational reasoning about the system functioning and a human-like believable dialogical interaction with the users.

François Bouchet & Jean-Paul Sansonnet (2011). Influence of Personality Traits on the Rational Process of Cognitive Agents. 2011 IEEE/WIC/ACM International Conferences on Web Intelligence and Intelligent Agent Technology.

In this paper we present an approach based on the principle that psychological capacities, especially personality traits, influence the decision making process of rational agents. While using the FFM/NEO PI-R taxonomy, we propose a model for the expression of personality traits in terms of so-called influence operators that add meta control rules to the cycle of rational BDI agents.

Jason M. Harley, François Bouchet & Roger Azevedo (2011). Examining Learners’ Emotional Responses to Virtual Pedagogical Agents’ Tutoring Strategies. Intelligent Virtual Agents, 11th International Conference.

What is the impact of a virtual pedagogical agent’s (VPA) tutorial strategy on learners’ emotions during the sub‐goal setting phase of their interaction with MetaTutor, an adaptive hypermedia learning environment? 18 undergraduate students (78% female) from two large, public universities were randomly assigned to one of two tutorial strategies deployed by a VPA: (1) Prompt Only Condition (PO), where VPA prompted students to set three sub‐goals, be mindful of their overall learning goal and either accepted or rejected students’ proposed sub‐goals, and (2) Prompt and Feedback Condition (FB), where VPA additionally provided information related to the relevancy and proximity of participants proposed subgoal to one or more of the seven ideal sub‐goals. Results indicated that the tutorial strategy deployed by the VPA had a significant impact upon the emotions participants experienced, in particular negative emotions. Specifically, participants embodied more negative emotions in the PO condition, anger in particular.

2010

François Bouchet & Jean-Paul Sansonnet (2010). Un cadre de modélisation des relations entre les réactions rationnelles et comportementales des agents assistants conversationnels. SMA Systèmes Multi-agents JFSMA 10 - 18e Journées francophones sur les systèmes multi-agents.

Afin d’améliorer l’acceptabilité des agents assistants conversationnels (AAC) auprès des utilisateurs, il est nécessaire de les munir de modèles comportementaux interagissant avec le processus de raisonnement rationnel de ces agents. Nous présentons un cadre de modélisation flexible, destiné à l’étude des relations entre les réactions rationnelles et comportementales d’AAC. Ce cadre est ensuite utilisé pour implémenter une première étude de cas, fondée sur la notion de biais cognitifs.

Marcelo Soares Pimenta, Evandro Manara Miletto, Jean-Paul Sansonnet & François Bouchet (2010). Social Music Making on the Web with CODES. Proceedings of the 2010 ACM Symposium on Applied Computing.

Music making is usually considered as mostly a solitary activity done by composers, but with the currentWeb 2.0 technology it is possible to provide new possibilities for social music making. CODES is a Web-based networked music environment designed to support music creation by novices in a cooperative and prototypical way, since no previous musical knowledge is required. Differently from others social media, where people only publish their content created elsewhere, in CODES novices can draft and refine cooperatively simple musical pieces, actually creating their own music, instead of only consuming it. This paper presents the main characteristics of CODES for social music making, with special focus on novices in music.

Jean-Paul Sansonnet & François Bouchet (2010). Joint handling of Rational and Behavioral reactions in Assistant Conversational Agents. Proc. of the 19th European Conference on Artificial Intelligence (ECAI 2010).

We describe here a framework dedicated to studies and experimentations upon the nature of the relationships between the rational reasoning process of an artificial agent and its psychological counterpart, namely its behavioral reasoning process. This study is focused on the domain of Assistant Conversational Agents which are software tools providing various kinds of assistance to people of the general public interacting with computer based applications or services. In this context, we show on some examples how the agents must exhibit both rational reasoning about the system functioning and a human-like believable dialogical interaction with the users.

François Bouchet & Jean-Paul Sansonnet (2010). Implementing personality adjectives as behavioral schemes. Proc. of Interfaces and Human-Computer Interaction 2010.

In this paper we present a methodology dedicated to the computational implementation of personality traits in Conversational Agents. First, a significant set of personality-traits adjectives is registered from thesaurus sources. Then the lexical semantics related to personality-traits is extracted while using the WordNet database and it is given a formal representation in terms of so-called Behavioral Schemes. Finally, we propose a framework for the implementation of those schemes as influence operators controlling the decision process and the plan/action scheduling of a rational agent.

Jean-Paul Sansonnet & François Bouchet (2010). Extraction of Agent Psychological Behaviors from Glosses of WordNet Personality Adjectives. Proc. of the 8th European Workshop on Multi-Agent Systems (EUMAS'10).

Conversational agents have two main parts: the rational agent performs symbolic reasoning over the model of the system while the psychological agent is in charge of the interaction with the user. In this context, agent cognitive modeling focuses on the taxonomy and on the computational implementation of psychological behaviors, in close relation with the rational reasoning. We present here the process of elicitation of a large class of psychological behaviors for their computational implementation in conversational agents. Behaviors are extracted from a corpus of personality adjectives associated with synsets and glosses in the WordNet lexical data base. Collected glosses are classified along the standard FFM - NEO PI-R taxonomy of personality traits in order to constitute clusters, now available as a resource for the community of conversational agent modeling.

Jean-Paul Sansonnet & François Bouchet (2010). Expression of behaviors in Assistant Agents as influences on rational execution of plans. Intelligent Virtual Agents, 10th International Conference.

Assistant Agents help ordinary people about computer tasks, in many ways, thanks to their rational reasoning capabilities about the current model of the world. However they face strong acceptability issues because of the lack of naturalness in their interaction with users. A promising approach is to provide Assistant Agents with a personality model and allow them to achieve behavioral reasoning in conjunction with rational reasoning. In this paper, we propose a formal framework to study the relationships between the rational and behavioral processes, based on the expression of the behaviors in terms of influence operators on the rational execution of actions and plans.

François Bouchet & Jean-Paul Sansonnet (2010). Classification of Wordnet Personality Adjectives in the NEO PI-R Taxonomy. Actes du Quatrième Workshop sur les Agents Conversationnels Animés (WACA 2010).

This work describes the process of classification of a set of personality-trait adjectives within the facet list of the NEO PI-R taxonomy related to the Five Factor Model. The classification process is not only based on adjective words but primarily on their lexical semantics as it is expressed by the synset-gloss attached to the adjectives in the Wordnet lexical base. This classification will provide a good coverage and support for the study and the computational implementation of psychological behaviors in conversational agents.

2009

François Bouchet & Jean-Paul Sansonnet (2009). Tree-kernel and Feature Vector Methods for Formal Semantic Requests Classification. Machine Learning and Data Mining in Pattern Recognition.

In this paper, we're interested in the classification of natural language requests converted into a formal representation, where the classes represent the conversational activity of those requests. This study is based on a corpus of requests collected using an assisting conversational agent, in which we identified four different classes (control, chat, direct and indirect assistance requests). The objective would be to take over from the rule-based system when it fails. First representing formal requests as a tree, we show it is possible to adapt tree kernel methods to our problematic. A second approach consisting in ignoring the request structure to focus on its components (i.e. considering it as a feature vector) gives better results a priori. We finally consider combining several of the previous classifiers, thus reaching a performance rate of 76.1%, which could be enough for using it as a complementary system.

François Bouchet & Jean-Paul Sansonnet (2009). Subjectivity and Cognitive Biases Modeling for a Realistic and Efficient Assisting Conversational Agent. 2009 IEEE/WIC/ACM International Conference on Web Intelligence and Intelligent Agent Technology.

Conversational agents are a promising way to provide assistance to novice users. After a semantic analysis, natural language requests are transformed into a formal representation the agent is using in conjunction with a model of the application to define the most appropriated reaction. But heuristics associating behaviors to patterns of semantically similar requests often fail to provide a reaction both efficient and realistic when they are only based on purely rational decisions. Therefore, we propose here an architecture for assisting conversational agents based on two notions: heuristics taking into account both rational and subjective parameters (based on a psychological model of the agent), and biases used to model deep personality contraints the agent can’t modify (implemented as modifiers over the messages transmitted by the agent). We illustrate its functioning with typical requests extracted from a corpus of requests to an assisting agent.

Evandro Manara Miletto, Marcelo Soares Pimenta, François Bouchet, Jean-Paul Sansonnet & Damián Keller (2009). Music Creation by Novices should be both Prototypical and Cooperative - Lessons Learned from CODES. Proc. of the 12th Brazilian Symposium on Computer Music (SBCM).

Musical creation is usually considered as mostly a solitary activity done by composers but we are convinced that CODES - a Web-based environment designed to support Cooperative Music Prototyping (CMP) - offers great contributions to social ways of music creation by novices. One of the main findings obtained during CODES development and usage is that systems aiming at providing effective support to such music creation activities by novices should meet specific requirements, in order to support a dynamic and creative environment, enabling knowledge sharing by means of rich interaction and cooperative mechanisms adapted to address the idiosyncrasies of this CMP context. The goal of this paper is to present, discuss and illustrate these prototypical and cooperative aspects of novice-oriented music creation activities in CODES.

Mao Xuetao, Jean-Paul Sansonnet & François Bouchet (2009). Intelligent Assisting Conversational Agents viewed through Novice Users' Requests. Proc. of Interfaces and Human-Computer Interaction 2009.

Assisting Conversational Agents are Embodied Conversational Agents dedicated to the Function of Assistance for applications and services to the general public, especially on the Internet. We have developed a web-based framework to experiment with assisting agents regarding the key issue of believability, and where good Natural Language Understanding is a primary concern. Now, we are confronted with the difficult issue of the cost of developing and customizing Natural Language Processing tools (NLP-tools) for each new assisted application. In this paper, we propose an approach which is a tradeoff between complex dialogue systems and naive chatbot systems. We think that our approach is worth considering because it focuses on a concise and well circumscribed linguistic domain: the domain of Assistance Requests that we captured by registering a corpus, in various contexts with ordinary people placed in front of Assisting Conversational Agents.

assisting conversational agents eliciting user requirements natural language processing

Mao Xuetao, François Bouchet & Jean-Paul Sansonnet (2009). Impact of agent's answers variability on its believability and human-likeness and consequent chatbot improvements. Proc. of the Symposium Killer Robots vs Friendly Fridges – The Social Understanding of Artificial Intelligence (AISB 2009).

Although globally less efficient than advanced dialogue systems, the chatbot approach allows people to easily design conversational agents. We suggest that one of their main drawbacks, their lack of believability, could be bypassed through the addition of variability in their answers, particularly when the variations depend on previous interactions or on particular parameters defining the agent. We validate the legitimacy of that hypothesis in two steps: first through simple additions to our chatbot-like framework (DIVA), we show it is technically feasible to simulate degrees of variability in answers. Then through an experiment done on 21 subjects interacting with two among six DIVA agents with different degrees of variability in a classical meeting scenario, we show that agents with an advanced variability in their answers are indeed perceived as more believable, human-like, and globally, more satisfying.

Jean-Paul Sansonnet, Evandro Manara Miletto, François Bouchet & Marcelo Soares Pimenta (2009). Exploring the integration of teaching capabilities into a CSCP framework through help agents. Proc. of the 20th Brazilian Symposium of Computer Science in Education (SBIE 2009).

In this paper we describe and we evaluate the first stages of a strategy for attaching basic teaching capabilities to Computer Supported Cooperative Prototyping (CSCP) environments. The strategy is based on the integration of learning functions in the reasoning module of an Embodied Conversational Agent in charge of the Help Function of the application. This paper presents a CSCP framework dedicated to music prototyping on the Web which enabled us to conduct a first experiment to evaluate the grounds of our strategy.

Evandro Manara Miletto, Jean-Paul Sansonnet, Marcelo Soares Pimenta & François Bouchet (2009). Corpus-based design of a Web 2.0 Assisting Agent. Proc. of the 8th International Workshop on Web-Oriented Software Technologies (IWWOST'2009).

We present an approach to facilitate the design of Assisting Conversational Agents dedicated to the Function of Assistance to ordinary people interacting in Natural Language with assisting agents on Web 2.0 pages. For each new assisted RIA, an experiment is carried out to collect a specific corpus of textual questions. It is then analyzed to exhibit the specific linguistic entities of the application required by the generic skeleton of the agent. Using a real large-scale example of a cooperative music prototyping application, we study the feasibility and evaluate the cost-effectiveness of this approach.

RIA assisting agents corpus-based design music prototyping

François Bouchet (2009). Characterization of Conversational Activities in a Corpus of Assistance Requests. Proc. of the 14th Student Session of the European Summer School for Logic, Language, and Information (ESSLLI).

Modeling linguistic interaction is a crucial point to provide assistance to ordinary users interacting with computer-based systems and services. A first issue is more particularly the characterization of the linguistic phenomena associated with the Function of Assistance, in order to define an assisting rational agent capable of pertinent reactions to users’ requests. In this paper, we present a corpus based on users’ requests registered in actual assisting experimentations. Using a method based on interactional profiles, we first compare the corpus with other similar dialogical corpora in order to assess its specificity, before focusing in a second time on characterizing its main linguistic features and the involved conversational activities.

assistance requests corpus conversational activities interactional profiles

François Bouchet & Jean-Paul Sansonnet (2009). Apports Complémentaires de la Subjectivité et des Biais Cognitifs à la Rationalité dans le Contexte de la Fonction d’Assistance. Actes des Cinquièmes Journées Francophones Modèles Formels de l'Interaction (MFI'09).

Les agents conversationnels sont un moyen prometteur d’assister des utilisateurs novices. Après une analyse sémantique, les requêtes en langue naturelle sont transformées en une représentation formelle utilisée en conjonction avec le modèle de l’application pour définir la réaction la plus appropriée. Cependant, les heuristiques associant des comportements à des schémas de requêtes sémantiquement similaires ne parviennent souvent pas à fournir une réaction efficace et réaliste quand elles se basent uniquement sur des décisions purement rationnelles. Nous proposons donc dans cet article une architecture d’agents conversationnels assistants fondée sur deux éléments : des heuristiques prenant en compte des paramètres à la fois rationnels et subjectifs (basés sur un modèle de personnalité de l’agent), et des biais utilisés pour modéliser des contraintes profondément liées à sa personnalité que l’agent ne peut modifier. Nous illustrons son fonctionnement sur des requêtes typiques issues d’un corpus de requêtes collectées avec un agent assistant. — Conversational agents are a promising way to provide assistance to novice users. After a semantic analysis, natural language requests are transformed into a formal representation the agent is using in conjunction with a model of the application to define the most appropriated reaction. However, in many cases, heuristics associating behaviors to patterns of semantically similar requests fail to provide a reaction both efficient and realistic when they are only based on purely rational decisions. To face this issue, we propose in this article an architecture for assisting conversational agents based on two elements : heuristics taking into account both rational and subjective parameters (based on a psychological model of the agent), and biases used to model deep personality contraints that the agent is unable to modify. We illustrate its functioning over some typical requests extracted from a collected corpus of requests to an assisting agent.

agent conversationnel assistance heuristiques personnalité biais cognitifs conversational agent assistance heuristics personality cognitive bias

Mao Xuetao, François Bouchet & Jean-Paul Sansonnet (2009). An ACA-based Semantic Space for Processing Domain Knowledge in the Assistance Context. Proc. of the 3rd International Conference on New Trends in Information and Service Science (NISS 2009).

For decades, people have been willing to interact with embodied conversational agents. This has driven researchers to consider more about social engineering than pure technical programming in building agent intelligence. As a typical application, assisting conversational agents aimed at helping people with attempting web-based applications have seen efficiency in many areas from e-learning to e-business and on-line games. The most outstanding feature of this kind of applications is that the assisting agent plays a role as a mediator conversing with users regarding the profile or topic of the to-be-assisted application. In this paper, we present a semantic space to model the knowledge in the assistance context considering both static and dynamic properties of the helping systems.

agent ACA ECA CHS semantic space

François Bouchet & Jean-Paul Sansonnet (2009). A framework for modeling the relationships between the rational and behavioral reactions of assisting conversational agents. Proc. of the 7th European Workshop on Multi-Agent Systems (EUMAS'09).

In order to facilitate the access and the usage to the general public of the rapidly expanding applications and services, particularly on the Internet, new assisting tools are needed with two main requirements: naturalness and acceptability. Conversational Agents are a promising approach for the support of the Function of Assistance, especially when they focus on the Natural Language modality. In this context, the assisting agents cannot rely only on rational reasoning over the structure and the functioning of the assisted application in order to resolve the user's questions. Agents must also express behavioral reactions that involve social relationships, character traits and affects. Once studied in separate communities, the relationships between rational and behavioral reactions are now considered a key issue. Some models have been proposed where the relationships are preset and often rigid while we think that more flexible tools would be handy to make experimental studies of this problem. In the first part of this paper, we propose a flexible framework for modeling the relationships between the rational and behavioral reactions of an assisting agent. Then this framework is used to support a first case-study, based on cognitive biases.

2008

Mao Xuetao, Jean-Paul Sansonnet & François Bouchet (2008). A corpus-based NLP-chain for a web-based Assisting Conversational Agent. Actes du Troisième Workshop sur les Agents Conversationnels Animés (WACA 2008).

Assisting Conversational Agents are Embodied Conversational Agents dedicated to the Function of Assistance for applications and services to the general public. Assisting agents for the general public are more and more required on the Internet-based new rich-client applications. We have developed a web-based framework to experiment with assisting agents, called the DIVA toolkit, where the Function of Assistance is a key issue, and the Natural Language modality a primary concern. This is why the DIVA toolkit is based on a Natural Language Processing chain (NLP-chain) handling the users’ textual questions about the structure and the functioning of the DIVA web pages. This paper describes the architecture of the NLP-chain and focuses on the corpus-based approach developed so as to provide an actual grounding for the intermediate Formal Request Form (FRF), at the heart of the NLP-chain.

web-based agents corpus of assisting requests natural language request handling

2007

François Bouchet (2007). Caractérisation d’un Corpus de Requêtes d’Assistance. Actes de la 11e Rencontre des Etudiants Chercheurs en Informatique pour le Traitement Automatique des Langues (RECITAL 2007).

Afin de concevoir un agent conversationnel logiciel capable d’assister des utilisateurs novices d’applications informatiques, nous avons été amenés à constituer un corpus spécifique de requêtes d’assistance en français, et à étudier ses caractéristiques. Nous montrons ici que les requêtes d’assistance se distinguent nettement de requêtes issues d’autres corpus disponibles dans des domaines proches. Nous mettons également en évidence le fait que ce corpus n’est pas homogène, mais contient au contraire plusieurs activités conversationnelles distinctes, dont l’assistance elle-même. Ces observations nous permettent de discuter de l’opportunité de considérer l’assistance comme un registre particulier de la langue générale. — In order to conceive a conversational agent able to assist ordinary people using softwares, we have built up a specific corpus of assistance requests in french, and studied its characteristics. We show here that assistance requests can be clearly distinguished from the ones from other available corpora in related domains. We also show that this corpus isn’t homogenous, but on the contrary reflects various conversational activities, among which the assistance itself. Those observations allow us to discuss about the opportunity to consider assistance as a general language particular registre.

corpus de requêtes d’assistance agent conversationnel activité conversationnelle actes de dialogue corpus of assistance requests conversational agent conversational activity speech acts

François Bouchet & Jean-Paul Sansonnet (2007). Caractérisation de Requêtes d’Assistance à partir de corpus. Actes des Quatrièmes Journées Francophones Modèles Formels de l'Interaction (MFI'07).

La modélisation formelle de l’interaction entre les usagers grand public et les systèmes informatiques a un rôle crucial à jouer au niveau de la Fonction d’Assistance. Il s’agit préalablement de caractériser sémantiquement, en termes de couverture et de précision, des phénomènes liés à l’assistance pour proposer à terme un agent rationnel assistant générique capable de réactions pertinentes aux requêtes des usagers. Dans cet article, nous présentons notre approche de construction d’un langage pour des agents conversationnels assistants, basée sur l’étude préalable d’un corpus de requêtes recueillies dans des situations effectives d’assistance. — Formal modeling of the interaction between ordinary users and computer-based systems has a major part to play in the Assistance Function. A first objective is to characterize semantically, both in coverage and precision, the phenomena associated with the Assistance Function to provide a generic assisting rational agent capable of pertinent reactions to users’ requests. In this paper, we present our approach to the construction of a language for a class of assisting conversational agents, based on the study of a corpus of users’ requests registered in actual assisting experimentations.

agents conversationnels assistants corpus de requêtes d’assistance langage de requêtes assisting conversational agents assistance requests corpus requests language

2006

François Bouchet & Jean-Paul Sansonnet (2006). Étude d'un corpus de requêtes en langue naturelle pour des agents assistants. Actes du Deuxième Workshop sur les Agents Conversationnels Animés (WACA 2006).

Un agent conversationnel peut être utilisé pour assister des utilisateurs novices d’applications informatiques. Pour réaliser un système adaptable, nous proposons dans cet article de le fonder sur un corpus spécifiquement constitué pour cet objectif. Nous justifions de la nécessité de construire un tel corpus, qui se distingue des corpus classiques, et détaillons la manière dont nous l’avons construit. Après avoir vérifié son adéquation par rapport à nos objectifs, nous cherchons à en dégager les spécificités en vue de concevoir un langage de requêtes formelles adapté à la fonction d’assistance dans les agents conversationnels. — A conversational agent can be used to assist ordinary people using softwares. To devise an adaptable system, we suggest in this paper to base it on a corpus built up in this objective. We justify the need for such a corpus, different from conventionnal corpora, and detail the way it has been built. After a check of its appropriateness to our objectives, we try to draw its particularities in order to create a fitting formal requests language for the assistance function in conversationnal agents.

agents assistants fonction d’assistance corpus de requêtes actes de dialogue assisting agents assistance function corpora of requests speech acts

Theses

2021

Camila Canellas (2021). Métamodèle d'analytique des apprentissages avec le numérique. Sorbonne université — PhD thesis.

Ce travail s’inscrit dans une démarche d’implémentation d’un processus d’analytique des apprentissages avec le numérique dans un contexte où la production documentaire est réalisée via une approche d’ingénierie dirigée par les modèles. Nous nous intéressons principalement aux possibilités qui pourraient émerger si une même approche est utilisée afin de réaliser une telle implémentation. Notre problématique porte sur l’identification de ces possibilités, notamment en s’assurant de permettre, via le métamodèle proposé, l’enrichissement d’indicateurs d’apprentissage avec la sémantique et la structure des documents pédagogiques consultés par les apprenants, ainsi qu’une définition en amont des indicateurs pertinents. Afin de concevoir le métamodèle en question, nous avons d’un côté procédé à une étude exploratoire auprès des apprenants afin de connaître leurs besoins et la réception d’indicateurs enrichis. D’un autre côté, nous avons réalisé une revue systématique de la littérature des indicateurs d’interaction existants dans le domaine de l’analytique des apprentissages avec le numérique afin de connaître les éléments potentiellement à abstraire pour la construction d’un métamodèle qui les représente. L’enjeu a été celui de concevoir un métamodèle où les éléments nécessaires à l’abstraction de ce domaine soient présents sans être inutilement complexes, permettant de modéliser à la fois des indicateurs d’apprentissage se basant sur une analyse descriptive et ceux faisant une prévision ou un diagnostic. Nous avons ensuite procédé à une preuve de concept et à une évaluation de ce métamodèle auprès des modélisateurs.

Learning analytics Sémantique 006.3 Analyse des données Analytique des apprentissages Apprentissage – Modèles mathématiques IDM Indicateurs MDE Metamodel Métamodèle Structure Technologie éducative Traces numériques

2019

Guy Merlin Mbatchou Nkwetchoua (2019). Vers un modèle d'accompagnement de l'apprentissage dans les Learning Management Systems : une approche basée sur la modélisation multi-scénarios d'un cours et la co-construction du scénario par les apprenants. Sorbonne université — PhD thesis.

Cette thèse contribue à l'accompagnement de l'apprentissage dans des EIAH dans le but d’améliorer le processus d'apprentissage. Dans un contexte où nous ne disposons pas des profils, comment accompagner en adaptant l’apprentissage ? Nous optons pour un apprentissage dirigé par l’apprenant sous les contraintes définies par l'enseignant. Nous avons élaboré un modèle de conception multi-scénarios du cours à destination des enseignants inspiré de la Competence-based Knowledge Space Theory à laquelle 3 extensions (activités à objectifs multiples, contraintes temporelles et seuils de satisfaction) sont ajoutées pour corriger ses faiblesses dans un contexte de formation initiale ou formation tout au long de la vie. Le modèle se base sur les objectifs d'apprentissage et les relations de précédence entre eux pour produire plusieurs scénarios en un temps raisonnable. Un sondage auprès des enseignants montre a priori une acceptabilité du modèle. Pour accompagner l'apprenant, nous lui offrons la possibilité de co-construire son scénario durant l'apprentissage. La co-construction résulte du respecter des contraintes définies par l'enseignant pour éviter des choix illogiques pouvant conduire à l'échec voire l'abandon. Le processus d’apprentissage se base sur le choix et changement des objectifs à atteindre et des activités à faire. Un sondage auprès des apprenants montre a priori une acceptabilité du modèle. Les modèles sont implémentés sous forme de plugin intégrable dans Moodle. Une expérimentation auprès des enseignants leur a permis de déceler des incohérences et des insuffisances dans leurs cours. Nous avons observé une diversité de scénarios construits par les étudiants.

004 006 Accompagnement de l'apprentissage Apprentissage dirigé par l'apprenant Apprentissage interactif Co-construction du scénario Co-construction of learning scenario Conception centrée sur l'utilisateur Conception multi-scénarios Environnements informatiques pour l'apprentissage humain Knowledge space theory Learner-directed learning Moodle (logiciel) Multi-scenario design Plateformes d'apprentissage en ligne Support of learning Technology enhanced learning Théorie des espaces de connaissance

Fatima Harrak (2019). Analyse de questions d’apprenants et de profils associés dans des environnements en ligne. Sorbonne université — PhD thesis.

Les questions des élèves sont utiles pour leur apprentissage et l'adaptation pédagogique des enseignants. Cependant, le volume de questions posées en ligne par les étudiants peut empêcher les enseignants de traiter chaque question (e.g. MOOC ou large cohorte universitaire). Nous abordons cette problématique principalement dans le cadre d’une formation hybride dans lequel chaque semaine les étudiants posent des questions en ligne, selon une approche de classe inversée, pour aider les enseignants à préparer leur séances de questions-réponses en présentiel. Notre objectif est d’outiller l’enseignant pour qu’il détermine les types de questions posées par les différents groupes d’apprenants. Pour mener ce travail, nous avons développé un schéma de codage de questions guidé par l’intention des élèves et la réaction pédagogique de l’enseignant. Plusieurs outils de classification automatique ont été conçus, évalués et combinés pour catégoriser les questions. Nous avons montré comment un modèle dérivé de clustering des données et entraîné sur des sessions antérieures peut être utilisé pour prédire le profil des élèves en ligne et établir des liens avec leurs questions. Ces résultats nous ont permis de proposer trois organisations de questions aux enseignants (basées sur les catégories de questions et profils des apprenants) qui ouvrent des perspectives de traitement différent lors des séances de questions-réponses. Nous avons testé et montré la possibilité d’adapter notre schéma de codage et les outils associés au contexte très différent d’un MOOC, ce qui suggère une certaine généricité de notre approche.

Blended learning Analyse des données 004 006.312 Automatic classification Classe inversée Classes inversées Classification automatique Clustering Enseignement – Méthodes actives Formation en ligne Mooc Profil d'étudiants Questions d'étudiants Students' profile Students' questions

2014

Issam Bani (2014). Analyse des traces d'apprentissage et d'interactions inter-apprenants dans un MOOC. Université Pierre et Marie Curie — MSc thesis.

Depuis leur apparition en 2008, les MOOCs (Massive Open Online Courses), connaissent un essor remarquable et prennent de plus en plus d’ampleur en matière d’enseignement supérieur. Cependant un problème majeur dans ce nouveau système éducatif est le taux élevé d’attrition. Nous nous sommes intéressés dans ce projet à rechercher une solution permettant de diminuer ce taux par l’utilisation de méthodes de Machine Learning. En premier lieu nous avons utilisé des algorithmes d’apprentissage supervisé pour déterminer les participants au MOOC en situation de décrochage. En deuxième lieu, tout en se basant sur le clustering, nous avons essayé de regrouper ces participants avec ceux qui sont dans les temps, sur la base d’informations issues de questionnaires individuels, de leurs habitudes de connexion à la plate-forme et de leurs interactions via le forum, pour créer des groupes afin d’assurer une entraide entre eux.

2010

François Bouchet (2010). Conception d'une chaîne de traitement de la langue naturelle pour un agent conversationnel assistant. Université Paris-Sud 11 — PhD thesis.

With the increasing number of novice users of computer applications, the need for efficient assistance has become critical. To supply the need, we suggest using an Assistant Conversational Agent (ACA), an interface allowing the use of natural language (used spontaneously when a problem arises) and providing a reassuring presence to the users. A preliminary study details the constitution (combining collection and the use of thesauri) of a corpus of requests, which need is justified. This corpus of 11,626 requests is compared with others, and we show that it covers the studied domain of assistance and moreover, contains requests regarding controlling of the application and chatting with the agent. This corpus provides a sound foundation for the conception of a syntactico-semantic analyzer of natural language requests, using a set of semantic keys, a set of analysis rules and a set of transformation rules. In output, requests are expressed in a formal language (DAFT) for which we provide the syntax and the semantics. The analyzer is evaluated by comparing a manual annotation and the requests automatically produced, and we consider the use of some supervised machine learning approaches in order to identify conversational activities. The methodology followed is validated through the integration of an ACA into an existing Web application for cooperative music prototyping. Finally, we describe the required architecture for the rational agent in charge of defining the reactions based on the formal requests expressed in DAFT and on the model of the assisted application, emphasizing the need for a cognitive model.

2006

François Bouchet (2006). Conception d'un langage de requêtes pour un agent conversationnel assistant. Université Paris-Sud 11 — MSc thesis.

Dans le cadre du projet DAFT qui a pour objectif la réalisation d'un agent conversationnel assistant capable de raisonnement sur la structure et le fonctionnement d'une application, nous cherchons à élaborer un langage de requêtes structuré dans lequel seront transformées les requêtes langagières des utilisateurs pour être ensuite exploitées par le module de raisonnement. La conception de cette représentation intermédiaire fondée sur l'identification d'actes de dialogue se base sur l'étude réalisée lors de ce stage d'un corpus de requêtes d'assistance recueilli à cet effet.

agents conversationnels fonction d'assistance actes de dialogue corpus de requêtes

Reports

2019

Marc Barzman, Mélanie Gerphagnon, Olivier Mora, Genevieve Aubin-Houzelstein, Alain Benard, Caroline Martin, George-Louis Baron, François Bouchet, Juliette Dibie-Barthelemy, Jean-Francois Gibrat, Simon Hodson, Evelyne Lhoste, Yann Moulier Boutang, Sébastien Perrot, Fabrice Phung, Christian Pichot, Mehdi Siné & Thierry Venin (2019). Transition numérique et pratiques de recherche et d’enseignement supérieur en agronomie, environnement, alimentation et sciences vétérinaires à l’horizon 2040. Inra (technical report).

La transition numérique bouleverse l'enseignement supérieur et la recherche publics. Les transformations des pratiques et des modes d'organisation, des relations entre les acteurs de l'écosystème autour de la recherche et de l'enseignement, et le sentiment d'accélération génèrent diverses images du futur - fantasmées ou plausibles. Dans l'enseignement supérieur, la transition numérique modifie les contenus, les outils et les méthodes pédagogiques. Des algorithmes, des intelligences artificielles et le développement de plateformes numériques transforment la relation enseignant-apprenant. La profusion de ressources pédagogiques en ligne renforce les possibilités d'autoformation et questionne la contribution du présentiel, de la formation initiale et des diplômes dans le parcours de chacun. Des algorithmes peuvent dorénavant définir les contenus et parcours de formation les plus adaptés aux apprenants. Les rôles des enseignants, des formateurs et des apprenants évoluent dans le contexte de parcours numériques individualisés et de fonctionnement au sein de communautés d'apprentissage en ligne. Le foisonnement des outils numériques peut soit favoriser l'accès généralisé à l'apprentissage, soit renforcer les inégalités. Face aux grands opérateurs privés, la place de l'enseignement supérieur et de la formation publics dans le marché de la connaissance est remise en question. La recherche se trouve également face à des opportunités et des défis inédits. Le numérique permet l'émergence de nouveaux métiers et de nouvelles façons de produire, de valider et de faire circuler la connaissance. Les données massives favorisent les approches basées sur la fouille de données tandis que les outils d'intelligence artificielle et de simulation transforment les pratiques de recherche. De nouveaux métiers apparaissent afin de traiter, analyser et gérer les données massives. Des collectifs de recherche se structurent en réseaux, incluant parfois la société civile. De nouvelles interactions apparaissent au sein d'écosystèmes de recherche et d'innovation en évolution. Les nouvelles capacités de communication permettent une organisation distribuée, mondialisée, centrée autour de clusters territoriaux ou de réseaux individuels labiles. Enfin, le numérique - dans l'agriculture numérique, la foodtech, l'e-santé, ou les soins vétérinaires connectées - transforme les objets d'étude. Face à ces changements, l'objectif de cette prospective est d'éclairer le débat sur les implications de la transition numérique pour la recherche et l'enseignement supérieur dans les sciences agronomiques, de l'environnement, de l'alimentation et vétérinaires à l'horizon 2040. Au-delà de ces domaines, cette initiative traite des questions pertinentes pour un plus large éventail d'acteurs concernés par le fonctionnement de la recherche, l'apprentissage et les modes de partage des savoirs, de l'enjeu des données dans l'économie numérique et de nouvelles relations entre la science et la société.

2010

Jean-Paul Sansonnet & François Bouchet (2010). A Formal Notation for the Elicitation of Emotions based on the OCC Model. LIMSI-CNRS (technical report).

The purpose of this paper is to explore the OCC model as described in [3], called OCC88 in the following, and the more recent version presented in [2], further called OCC02. On this basis, we propose a new classification of emotions based on the classes of procedural situations in which the emotions are elicited. First we present a formalization of the OCC88 model taken from [4]; then we propose a first transformation based on the formalization of the activities. In a second study we present a formalization of the OCC02 model and which is again classified in terms of activities. Finally both the OCC88 and OCC02 models of emotion are classified in a new taxonomy which focuses on the nature of the arguments of the emotions, hence revealing holes in the OCC taxonomy which are filled with new emotions.

2007

François Bouchet & Jean-Paul Sansonnet (2007). Characterization of Conversational Activities in a Corpus of Assisting Requests. LIMSI-CNRS (technical report).

As far as interaction between ordinary users and the new computer-based systems and services is concerned, modeling of the linguistic interaction will play a major role, especially with regard to the Function of Assistance. A first issue is then the semantic characterization, both in coverage and precision, of the phenomena associated with the Function of Assistance in order to provide an assisting rational agent capable of pertinent reactions to users’ requests. In this paper, we present a corpus based on users’ requests registered in actual assisting experimentations. Using a method based on interactional profiles, first we compare the corpus with other dialogical corpora in order to assess its specificity, then we try to characterize its main linguistic features and the conversational activities involved.

Assisting requests corpus conversational activities interactional profiles

2006

François Bouchet (2006). Spécifications du langage de requêtes DAFT 2.0. LIMSI-CNRS (technical report).