Publications by Type

Abstract. Les tableaux de bord d'apprentissage, dont le déploiement doit soutenir la prise de décision à toutes les étapes de la formation, continuent de faire face à des défis d'adoption. La littérature identifie plusieurs raisons à cette réticence, notamment l'absence de principes directeurs pour la conception et une participation insuffisante des parties prenantes. Pour y remédier, nous proposons un cadre de conception ainsi que deux outils pour supporter le processus de conception. Le cadre de conception, axé sur la création de sens pour étayer la prise de décision, guide la conception au travers d'un espace de conception exhaustif. Le premier outil propose une phase d'idéation participative et encourage la collaboration des différentes parties prenantes. Le second outil propose de simplifier le travail du développeur et d'encourager sa collaboration avec les parties prenantes. L'évaluation de ces outils repose sur des critères tels que la facilité d'utilisation, le soutien à la participation et l'expérience utilisateur, éclairant chaque étape du processus de conception. Les résultats montrent que ces outils soutiennent significativement la participation des utilisateurs finaux et l'expérience des développeurs, soulignant leur pertinence et leur efficacité. En favorisant l'adoption des tableaux de bord d'apprentissage par une conception centrée sur l'utilisateur, cette contribution met en lumière le potentiel de ces outils et l'importance des approches participatives dans le développement des technologies éducatives.

English abstract. Despite their potential, learning dashboards face challenges in terms of adoption. Research highlights a lack of design principles and insufficient stakeholder participation as key factors. To address these, we propose a design framework emphasizing sensemaking for decision making. Additionally, we advocate for a collaborative design approach, supported by tools for ideation and prototyping. These tools aim to involve users actively, promoting an inclusive design approach. Their evaluation, based on ease of use, support for participation, and user experience, illuminates each design stage. Results show these tools significantly support end-user participation and developer experience, underscoring their relevance and effectiveness.

Abstract. Learning Analytics Dashboards (LAD) deserve increasing attention, yet their adoption remains limited. Designing effective LAD is a difficult process, and LADs often fail in turning insights into action. We argue that providing explicit decision-making features in a participatory design process may help to develop LADs supporting action. We first examine how the decision-making process is reflected on LADs. Second, we review the literature to identify major design space dimensions and examine how to include decision-making features. Third, we propose the DEFLAD design framework to synthesise this review which provides explicit decision-making features in three dimensions: goal expression as a situation awareness level, visualisation and related interactions, as support of decision-making process. Fourth, we consider how this framework is involved through every stage of a human-centred design (HCD) process to express and manage such features. The main contribution of this paper is to provide a framework integrating the decision-making features in a participatory design process of LADs. Furthermore, we demonstrate the implementation of our proposals through the development of a card-based toolkit to assist in the ideation phase of participatory design, and present feedback from participants of a workshop utilising this tool as a proof of concept.

Abstract. This paper proposes a graph-based approach to determine the importance of each media in a multimedia document by expanding the traditional four-dimensional model with a dimension that captures the rhetorical relations between different media types. The proposed approach utilizes an algorithm to weight the media types based on their significance. The use of rhetorical structure theory enables the determination of the significance of each media type, making it useful for document adaptation, automatic composition, and automatic generation of summaries. The approach utilizes an extended orbits-based rhetorical structure that is a novel method for determining the importance of media types in multimedia documents. The proposed approach is effective in capturing the importance of each media type and can be utilized in a wide range of applications, making it a potential solution to the limitations of the traditional model. This research has implications for a range of applications, including document adaptation, automatic composition, and automatic generation of summaries.

Abstract. In this paper, we propose a set of rhetorical relations to support applications such as automatic summary generation and content adaptation of a multimedia document. These relations have been proposed in the context of an educational environment. These rhetorical relations are integrated and handled as part of the logical dimension of the multimedia document. The proposal of these relations is motivated by the need to take into account the particularities inherent to: (1) the composition, editing and presentation of a multimedia document and (2) the educational context. Indeed, multimedia documents in an educational context are very different from textual documents, for which automatic analysis and generation have led to the proposal of a set of commonly used rhetorical relations, as described in the work of Mann and Thompson. The study of this now more common context allowed us to go beyond the body of existing work to develop a more appropriate set of rhetorical relationships related to educational multimedia documents.

Abstract. Un document multimédia intègre des entités de base de nature statique (texte, images, graphiques et tableaux) et de nature dynamique (vidéo, sons et animations), qui suivent une certaine organisation temporelle et spatiale. Ces documents sont généralement modélisés par quatre dimensions : la dimension logique, la dimension temporelle, la dimension spatiale et la dimension hypermédia. Avec le temps, de nouveaux besoins sont apparus dans l'utilisation des documents multimédia, notamment l'adaptation des documents, la composition, la recomposition, et la génération de résumé automatique. Les informations contenues dans ces quatre dimensions se révèlent insuffisantes pour répondre à ces besoins. Dans ce travail, nous proposons d'étendre ce modèle à cinq dimensions, en définissant une nouvelle dimension : la dimension annotation. Cette dernière comportera des informations sur le document, ses entités, son auteur et ses lecteurs, qui contribueront à la prise en charge des applications précitées.

Abstract. A challenge that course authors face when reviewing their contents is to detect how to improve their courses in order to meet the expectations of their learners. In this paper, we propose an analytical approach that exploits learners' logs of reading to provide authors with insightful data about the consumption of their courses. We first model reading activity using the concept of reading-session and propose a new and efficient session identification. We then elaborate a list of indicators computed using learners' reading sessions that allow to represent their behaviour and to infer their needs. We evaluate our proposals with course authors and learners using logs from a major e-learning platform. Interesting results were found. This demonstrates the effectiveness of the approach in identifying aspects and parts of a course that may prevent it from being easily read and understood, and for guiding the authors through the analysis and review tasks.

Abstract. Providing high-quality courses is of utmost importance to drive successful learning. This compels course authors to continuously review their contents to meet learners' needs. However, it is challenging for them to detect the reading barriers that learners face with content, and to identify how their courses can be improved accordingly. In this paper, we propose a learning analytics approach for assisting course authors performing these tasks. Using logs of learners' activity, a set of indicators related to course reading activity are computed and used to detect issues and to suggest content revisions. The results are presented to authors through CoReaDa, a learning dashboard empowered with assistive features. We instantiate our proposals using the logs of a major European e-learning platform, and validate them through a study. Study results show the effectiveness of our approach providing authors with more awareness and guidance in improving their courses, to better suit learners' requirements.

Abstract. Hypervideos are hypermedia documents that focus on video content. While they have long been deployed using specialized software or even hardware, the Web now offers a ground for them to fit into standardized languages and implementations. However, hypervideo design also currently uses very specific models limited to a single class of documents, or very generic hypermedia models that may not appropriately express their specific features. In this article we describe such features, and we introduce CHM, an annotation-driven and component-based model to conceptualize hypervideos through a high level operational specification. An extensible set of high level components is defined to emphasize the presentation and interaction features modeling, while lower level components offer more flexibility and customization opportunities. Being annotation-based, the model promotes a clear separation between video content/metadata and their various potential presentations. We also describe WebCHM, an implementation of CHM with standard Web technologies that provides a general framework to experiment with hypervideos on the Web. Two examples are provided as well as a preliminary usage study of the model and its implementation to validate our claims and proposals.

Abstract. Dans les standards et les systèmes multimédia actuels, le document est le plus souvent décrit selon une approche événementielle, d'où l'inadaptation de la programmation à la nature incrémentale du processus d'édition, la difficulté de la maintenance des documents ainsi produits et enfin les problèmes qu'ont les auteurs non-informaticiens pour les maîtriser. De plus, les travaux menés dans ce domaine se sont essentiellement focalisés sur la dimension temporelle, alors que le développement de tels systèmes devrait d'emblée considérer les quatre dimensions du document afin de mutualiser au mieux leur conception et leur implémentation. Par ailleurs, l'adaptation des documents à des profils de lecture particuliers n'est pas considérée, alors que pour certains domaines d'application, l'apprentissage par exemple, ceci reste un besoin important. En partant de ce constat, MediaStudio se veut une contribution au domaine de l'édition et de la présentation de documents multimédia interactifs à travers un ensemble de propositions pour pallier les insuffisances mentionnées.

Résumé. L'accessibilité des contenus STEM pour les personnes aveugles ou malvoyantes se heurte à un paradoxe persistant : l'accroissement de la sophistication architecturale des interfaces multimodales ne réduit pas systématiquement la charge cognitive. Deux causes structurelles expliquent ce paradoxe. D'une part, une illusion additive traverse implicitement la conception~: accumuler des canaux sensoriels est confondu avec leur orchestration, c'est-à-dire la coordination temporelle et sémantique des flux d'information en vue d'une représentation mentale unifiée. D'autre part, l'évaluation repose sur des métriques de complétion de tâche insensibles aux coûts cognitifs sous-jacents. Nous introduisons ARC (Architectural Readiness for perceptual Convergence), un cadre diagnostique opérationnalisant la qualité de l'orchestration multimodale selon quatre dimensions architecturales - cohérence temporelle, cohérence informationnelle, préservation structurelle, pertinence modale - et une variable de performance externe. Appliqué à un corpus de 66 études empiriques (2015-2025) portant sur des interfaces STEM non visuelles, notre analyse descriptive identifie trois régimes de maturité architecturale qualitativement distincts. Malgré des écarts structurels majeurs entre régimes, les métriques de performance mesurées ne varient pas significativement, révélant un découplage systématique entre ces dernières et sophistication architecturale. Nous formalisons ce résultat sous l'hypothèse du Rendement Cognitif Différentiel : des systèmes architecturalement distincts produisent des performances équivalentes tout en imposant probablement des coûts cognitifs fondamentalement différents --- un construit théoriquement fondé, qui appelle une validation expérimentale par des protocoles multi-métriques.

Abstract. Multimodal STEM accessibility for blind and visually impaired users faces a persistent paradox: increasing architectural sophistication does not systematically reduce cognitive load. We argue that two structural causes underlie this gap. First, an additive illusion pervades design practice, equating the mere accumulation of sensory channels with their purposeful orchestration—that is, the temporal and semantic coordination of information streams into a unified mental representation. Second, evaluation relies predominantly on task-completion metrics that are insensitive to the underlying cognitive costs. We introduce ARC (Architectural Readiness for perceptual Convergence), a diagnostic framework operationalizing multimodal orchestration quality across four architectural dimensions—temporal synchrony, informational coherence, structural preservation, and modal relevance—plus one external performance variable. Applied to a corpus of 66 empirical studies (2015-2025) on non-visual STEM interfaces, our descriptive analysis identifies three qualitatively distinct regimes of architectural maturity. Despite large structural differences across regimes, performance metrics show no systematic variation, revealing a consistent decoupling between architectural sophistication and task success. We formalize this as the Differential Cognitive Yield (DCY) hypothesis: architecturally divergent systems yield equivalent observable performance while likely imposing fundamentally different cognitive costs—a construct requiring experimental validation with multi-metric protocols. ARC contributes a reproducible scoring protocol with codified inter-rater procedures, empirically derived classification thresholds, and hierarchical design guidelines.

Résumé. Les systèmes d'accessibilité fondés sur l'IA sont passés de la transcription (restituer tel quel le contenu visuel) à l'interprétation (descriptions sélectives, souvent conversationnelles). Ce changement, tout en enrichissant l'expérience, fait émerger un paradoxe de vérification : les déficients visuels (DV - aveugles ou malvoyants), qui dépendent le plus des descriptions de l'IA, n'ont plus d'accès perceptuel pour détecter des erreurs critiques (hallucinations, erreurs de classification, dérive sémantique). Cette asymétrie est consubstantielle au rôle d'intermédiaire épistémique de l'IA, qui façonne l'information tout en supprimant les moyens de la vérifier. À travers une revue systématique évaluant de manière comparative 17 systèmes, nous introduisons le Verification‑Cognition Balance Index (VCBI). Cette métrique composite évalue l'équilibre entre deux leviers de vérifiabilité que sont la rigueur de validation du système et l'agentivité épistémique (capacité des utilisateurs à détecter les erreurs du système), et deux sources de charge cognitive liée à la vérification : la complexité architecturale du système et la densité de risque documenté. Nos résultats montrent entre autres que les architectures basées sur les transformeurs présentent un risque documenté nettement supérieur à celui des pipelines de vision classique, sans amélioration correspondante de l'équilibre vérification‑cognition, et que les systèmes hybrides spécialisés, bien que rare, offrent une voie plus prometteuse. Surtout, les évaluations actuelles privilégient la mesure de la précision système au détriment de l'agentivité épistémique. Ancrés dans la théorie de l'agentivité épistémique, nous plaidons pour un recentrage de l'évaluation - de la précision système vers la vérifiabilité par l'utilisateur, et proposons trois schémas de conception actionnables—transparence de provenance, signalement non visuel de l'incertitude, et interactions de contestation—qui opérationnalisent ce principe pour les utilisateurs DV.

Abstract. AI-powered accessibility systems have shifted from transcription (rendering visual content as-is) to interpretation (generating selective, conversational narratives). While this evolution enriches user experience, it also introduces a verification paradox: blind and low-vision (BLV) users, who rely most heavily on AI descriptions, are the very people who lack the independent perceptual access needed to detect consequential errors—hallucinations, misclassifications, or semantic drift. This asymmetry is inherent whenever AI acts as an epistemic intermediary, shaping information while simultaneously removing the means to verify it. Through a systematic review with comparative analysis of 17 empirically evaluated systems, we introduce the Verification-Cognition Balance Index (VCBI). This composite metric quantifies the equilibrium between two enablers of verification (validation rigor and user agency) and two sources of epistemic burden (architectural complexity and documented risk density). Our findings reveal that transformer-based architectures exhibit substantially higher documented risk than classical computer-vision pipelines, yet they offer no corresponding improvement in verification balance. Specialized hybrid systems point toward a more promising, though presently rare, alternative. Most critically, current evaluations emphasize system accuracy while overlooking whether users can detect AI errors. Grounded in epistemic agency theory, we argue for a necessary shift from system accuracy to user verifiability and distill three actionable design patterns—provenance transparency, non-visual uncertainty signaling, and contestation interactions—that make this principle concrete for BLV users.

Abstract. Multimodal systems combining audio, haptic, and tactile channels are prevalent in STEM accessibility for blind and visually impaired users, yet real-world feedback reports high cognitive load despite technological advances. Through analysis of 66 systems (2015-2025), we identify three design regimes: additive (channel stacking), augmentative (partial coordination), and integrative (orchestrated fusion). We reveal a critical evaluation gap: while architectural quality varies dramatically, conventional performance metrics show no regime variation, indicating current metrics miss the cognitive cost of poorly integrated designs. We contribute: (1) a diagnostic framework with high reliability; (2) evidence of a performance-architecture decoupling where cognitive costs diverge despite similar task performance; and (3) design thresholds for creating perceptually integrated experiences. Our work calls for shifting from interface engineering to perceptual integration engineering.

Abstract. AI-powered accessibility systems have shifted from transcription to interpretation, creating a verification paradox: blind and low-vision (BLV) users, who most rely on AI-generated narratives, lack perceptual redundancy to detect errors such as hallucinations or semantic drift, resulting in an agency deficit. Through a systematic meta-analysis of 17 empirically evaluated AI accessibility systems (2019-2025), we examine how architectural choices shape verification affordances. We introduce the Verification-Cognition Balance Index (VCBI), a composite metric quantifying trade-offs between validation rigor, user agency, complexity, and risk. Our analysis uncovers weak overall balance across the field, with transformer-based systems showing higher risk density than classical computer vision despite negligible validation improvements. Only hybrid specialized architectures-constraining AI to narrow semantic subtasks-achieve good balance. We argue for a necessary paradigm shift in evaluation from system accuracy to user verifiability.

@inproceedings{sadallah2026w4a,
  author = {Sadallah, Madjid and Encelle, Beno\^{\i}t},
  title = {The Verification Gap in AI Accessibility: A Meta-Analysis of Architectural Paradigms and Agency Deficits},
  year = {2026},
  booktitle = {Proceedings of the 23rd International Web for All Conference (W4A'26)},
  pages = {211--221},
  publisher = {Association for Computing Machinery},
  address = {Dubai, United Arab Emirates},
  doi = {10.1145/3800424.3800451}
}

Abstract. Assistive educational technologies for blind and visually impaired STEM learners perform well in controlled studies yet fail to achieve sustained classroom adoption. Analysis of 25 empirical studies reveals the cause: tools are designed and evaluated against technical metrics that do not predict pedagogical success. We term this the technical capability fallacy: the assumption that technical capability ensures pedagogical utility. To address this, we propose FOCAL (Framework for Objective-Context ALignment), which specifies what a tool must provide for each combination of learning objective and educational context.

Abstract. La marginalisation du braille dans l'accessibilité numérique repose sur un paradigme de substitution modale postulant l'équivalence cognitive entre audio et accès tactile. Or, dans les domaines formels (mathématiques, programmation, sciences), cette équivalence s'effondre : les exigences de navigation non linéaire, de vérification syntaxique et de manipulation symbolique révèlent une complémentarité structurelle irréductible. Nous développons le modèle de l'Ancre et du Flux : l'Ancre (braille) assure stabilisation cognitive et audit épistémique, tandis que le Flux (audio-IA) permet exploration contextuelle et interaction fluide. Face aux hallucinations des IA génératives, le braille s'impose comme infrastructure critique de vérification indépendante. Nous établissons quatre principes normatifs de conception d'interfaces accessibles et soutenons que cette complémentarité constitue une nécessité architecturale de la cognition distribuée. Le droit à l'audit épistémique, matérialisé par l'accès braille, fonde ainsi la souveraineté cognitive à l'ère algorithmique.

Abstract. Les personnes déficientes visuelles (DV) restent marginalisées dans les parcours scientifiques, freinées par l'inaccessibilité des documents techniques et par l'absence de collaboration "équitable" avec les voyants. Si la conception participative est désormais une pratique courante, elle n'engage pas les utilisateurs DV comme co-chercheurs. Cet article présente l'application du partenariat épistémique (co-construction de savoirs) dans le cadre d'un projet ANR SAPS (Science Avec et Pour la Société) lié à l'accessibilité numérique. Notre contribution est méthodologique : nous opérationnalisons ce partenariat avec un niveau élevé d'exigence (constitution d'un collectif paritaire, gouvernance par non-dissensus, accessibilité intégrale du dispositif de recherche) ; nous démontrons la faisabilité de travaux empiriques rigoureux entièrement co-construits ; nous exploitons la réflexivité comme levier de conception, en transformant les difficultés d'awareness vécues dans le collectif en spécifications techniques. Cette approche renouvelle ainsi la recherche en accessibilité en répondant à l'ambition d'une science avec et pour la société.

Abstract. Notre proposition s'appuie sur le projet Mobiles qui s'attache à analyser la socialisation langagière d'étudiants internationaux en séjour d'immersion en France. L'hypothèse forte de cette recherche est qu'il convient de s'intéresser à l'hétérogénéité des pratiques socio-spatiales urbaines pour comprendre les processus de socialisation langagière. La socialisation des étudiants internationaux, immergés dans un nouveau contexte à la fois social et langagier est au cœur de notre problématique et nous cherchons à comprendre et analyser la matérialité et la spatialité de ces divers processus. La socialité s'instaure à travers les expériences partagées par les individus au sein d'espaces communs (Lussault, 2007), de communautés discursives et de pratiques (Duff, 2007). Considérant que l'espace est une performance collective (Besse, 2018), nous avons mis en place des formes et des situations collectives d'enquête pour comprendre les expériences des lieux et la manière dont la socialisation y était liée (Danos & Grassin, 2024). De plus, tout espace est l'espace d'une enaction identitaire (Moje, 2004) et celui d'un apprentissage expérientiel (Lovett, 2020). Ces situations d'enquête nous ont permis de recueillir des « small stories » (Bamberg & Georgakopoulou, 2008), récits courts, fragmentés et interactifs, qui mettent en évidence des orientations narratives vers le monde sensibles aux interactions avec l'environnement. Après avoir présenté ces situations d'enquête socio-spatiale, nous présentons les ressources pédagogiques créées par le projet : 1/ des ateliers de cartographie ont été conçus pour la classe de langue autour de grandes cartes de la ville de Lyon (format A1) ; 2/ une application de cartographie participative fonctionnant comme un architexte propice au récit multimodal, permettant l'écriture de l'expérience entre un « montrer » et un « raconter » (Grassin & Danos, 2024). En s'appuyant sur la participation et sur la dimension expérientielle de l'apprentissage, le dispositif relève d'une écologisation du design pédagogique (Paquelin, 2020) cherchant à favoriser l'engagement des sujets dans leur environnement et leur agentivité.

Abstract. L'intégration des étudiants internationaux, cruciale pour leur inclusion sociale et leur apprentissage, rencontre souvent divers obstacles. Bien que les outils numériques, telles les applications de découverte et de documentation, puissent soutenir ce processus, leur efficacité repose sur une appropriation réfléchie. Cet article présente un système de recommandations personnalisées conçu pour guider les étudiants internationaux dans l'utilisation d'une application d'exploration urbaine et de partage d'expériences. En offrant des suggestions contextuelles, ce système encourage une exploration active et une meilleure compréhension de l'environnement local, renforçant leur intégration sociale et culturelle. Une évaluation empirique a révélé une amélioration de la qualité narrative, de la compréhension du contexte local et de l'expression émotionnelle des étudiants. Ces résultats soulignent le potentiel des systèmes de recommandations pour enrichir l'apprentissage contextuel et faciliter l'intégration, facteurs clés du bien-être et de la réussite académique des étudiants internationaux.

Abstract. L'intégration des étudiants internationaux, cruciale pour leur réussite, se heurte souvent à des défis majeurs : isolement social dû au manque de repères et de réseaux, difficultés de progression linguistique limitant les interactions, et besoin d'engagement actif dans leur nouvel environnement. Pour tenter d'y répondre, nous présentons MOBILES, une application transformant l'exploration urbaine en un outil d'apprentissage et d'intégration. Elle incite les étudiants à explorer activement leur ville, à documenter et partager leurs expériences via des annotations géolocalisées, facilitant ainsi leur adaptation socioculturelle et linguistique informelle. Grâce à des recommandations personnalisées basées sur leurs intérêts et activités, MOBILES vise à renforcer leur sentiment d'appartenance et à stimuler leur engagement, créant un lien entre intégration et apprentissage situé. Cette démonstration présente les fonctionnalités clés de l'application.

Abstract. The question of 'place' remains a topical one in the field of Geographic Information Sciences, where 'space' is still predominant. This issue is explored here in the context of an experiment involving international students who have been asked to share their located experiences via a dedicated mobile application. We present the objectives, design and implementation of this application and discuss the issues associated with these observations in the field of Geographic Information Sciences.

Abstract. The integration of international students, crucial for their social inclusion and learning, often faces various obstacles. While digital tools, such as discovery and documentation applications, can support this process, their effectiveness depends on thoughtful adoption. This article introduces a personalized recommendation system designed to guide international students in using an urban exploration and experience-sharing application. By offering contextual suggestions, this system encourages active exploration and a deeper understanding of the local environment, thereby enhancing their social and cultural integration. An empirical evaluation revealed improvements in students' narrative quality, local context comprehension, and emotional expression. These findings highlight the potential of recommendation systems to enrich contextual learning and facilitate integration, key factors for the well-being and academic success of international students.

Abstract. Learning path recommendation systems typically rely on predefined knowledge structures, limiting their adaptability to learners' actual needs. In this work, we propose an approach that infers a learning space from learners' interaction data and uses it to produce learning path recommendations. Building on Knowledge Space Theory (KST), the inferred learning space encodes prerequisite relationships among learning objects, enabling structured and personalised guidance. The inferred structure is exploited to recommend next learning steps that are both coherent with the learner's current knowledge state and aligned with the overall learning objectives. We evaluate the proposed approach and discuss its potential to improve the relevance and adaptability of learning path recommendations.

Abstract. Les tableaux de bord d'apprentissage visent à soutenir la prise de décision au cours du processus d'apprentissage. Cependant, leur adoption à grande échelle reste limitée. Si ce constat peut s'expliquer par leur apparition relativement récente, la recherche montre que ceci est également dû au manque de participation des parties prenantes au processus de conception. Afin de soutenir ce processus tout en impliquant les utilisateurs, nous proposons d'articuler autour d'un espace de conception partagé un outil de conception participative pour la phase d'idéation et un outil génératif pour la phase de prototypage.

Abstract. Le concept d'orchestration de la classe vise à saisir la complexité associée à la gestion des activités d'apprentissage synchrones (où apprenants et enseignants se retrouvent pour réaliser les activités). Notre objectif est de construire un dispositif permettant d'assister l'enseignant dans ses tâches d'orchestration, en explorant les notions de scénarisation, de suivi et de rétroaction. Dans cet article, nous présentons SAVAStudio, un outil permettant à l'enseignant de planifier le déroulement d'activités synchrones et de récupérer les traces du déroulement réel de celles-ci.

English abstract. Classroom orchestration aims at capturing the complexity associated with managing synchronous learning activities (where learners and teachers work together to complete activities). Our goal is to build a tool to support the teacher in his orchestration tasks, by exploring the notions of scripting, monitoring and feedback. In this article, we present SAVAStudio, a tool that allows the teacher to plan the conduct of synchronous activities and to retrieve traces of their actual execution.

Abstract. Learning Analytics Dashboards (LADs) deliver rich and actionable representations of learning data to support meaningful and insightful decisions that ultimately leverage the learning process. Yet, because of their limited large-scale adoption, their design is still a major area of inquiry in education research. In this paper, we propose to expand LAD codesign approaches. We first investigate how the user makes sense of the data delivered by LADs and how to support this sensemaking process during the design. Second, we propose a generative tool, supporting sensemaking and decision making process, that extends end-users participation during the prototyping phase and empowers LAD designers.

Abstract. Learning Analytics Dashboards (LADs) are data visualization tools built to empower teachers and learners to make purposeful decisions that impact the learning process. Due to their relatively recent emergence and the scarcity of studies on their design principles, dashboard design remains a major area of investigation in learning analytics research, and large scale diffusion to their stakeholders remains limited. We promote human-centered approaches for LADs design since their success in terms of acceptance and adoption greatly depends on the level of stakeholder involvement in their design. In this paper, we present a tool to support the participatory design of LADs. First experiments during a pilot study with teachers demonstrate that the proposed tool encourages group work, and in-depth exploration of LADs use.

Abstract. It is a challenging task to identify eLearning courses parts that have to be revised to best suit learners' requirements. Reading being one of the most salient learning activities, one way of doing so is to study how learners consume courses. We intend to support course authors (e.g. teachers) during courses revision by providing them with reading indicators. We use the concept of reading session to denote a learner's active reading period, and we provide several associated reading indicators. In our server-side approach, reading sessions and indicators are calculated using web server logs. We evaluate the relevance of our proposals using logs from a major French eLearning platform. Results are promising: calculated reading sessions are theoretically more precise than other best applicable approaches, and course authors consider suggested indicators to be appropriate to courses revision. Using reading sessions and associated indicators could facilitate authors' work of course reengineering.

Abstract. This ongoing work investigates usage-based document reengineering as a means to support authors in modifying their documents. Document usages (i.e. usage feedbacks) cover readers' explicit annotations and their reading traces. We first describe a conceptual framework with various levels of assistance for document reengineering: indications on reading, problem detection, reconception suggestions and automatic reconception propositions, taking our example in e-learning document management. We then present a technical framework for usage-based document reengineering and its associated models for documents, annotations and traces representation.

Abstract. Hypervideo offers enhanced video-centric experiences. Usually defined from a hypermedia perspective, the lack of a dedicated specification hampers hypervideo domain and concepts from being broadly investigated. This article proposes a specialized hypervideo model that addresses hypervideo specificities. Following the principles of component-based modeling and annotation-driven content abstracting, the Component-based Hypervideo Model (CHM) that we propose is a high level representation of hypervideos that intends to provide a general and dedicated hypervideo data model. Considered as a video-centric interactive document, the CHM hypervideo presentation and interaction features are expressed through a high level operational specification. Our annotation-driven approach promotes a clear separation of data from video content and document visualizations. The model serves as a basis for a Web-oriented implementation that provides a declarative syntax and accompanying tools for hypervideo document design in a Web standards-compliant manner.

Abstract. Effective video-based Web information system deployment is still challenging, while the recent widespread of multimedia further raises the demand for new online audiovisual document edition and presentation alternatives. Hypervideo, a specialization of hypermedia focusing on video, can be used on the Web to provide a basis for video-centric documents and to allow more elaborated practices of online video. In this paper, we propose an annotation-driven model to conceptualize hypervideos, promoting a clear separation between video content/metadata and their various potential presentations. Using the proposed model, features of hypervideo are grafted to wider video-based Web documents in a Web standards-compliant manner.

Abstract. This protocol presents a registered systematic review that aims to synthesize the landscape of interactive digital technologies enabling blind and low-vision (BLV) individuals to access two-dimensional STEM content — including data visualizations, mathematical notations, diagrams, maps, and tables — through non-visual modalities (tactile, auditory, haptic, or multimodal). Accessibility research in STEM faces persistent methodological challenges: small sample sizes, high heterogeneity of study designs, and the socio-technical complexity of interventions, which often render traditional frameworks such as PICO inadequate. To overcome these limitations, this protocol introduces a novel and coherent integration of three complementary frameworks: PRISMA 2020 for rigorous and transparent reporting, SPIDER for conceptually sound organization of diverse evidence (particularly qualitative and mixed-methods studies), and MMAT 2018 for non-hierarchical quality assessment across five study typologies. The review further adopts Cognitive Load Theory to analyze the cognitive demands imposed by different access solutions and the Technology Acceptance Model (TAM) extended by UTAUT to examine usability, acceptance, and adoption factors. Methodological innovations include a hierarchical coding scheme with precise operational definitions, a dual-template data extraction architecture (broad mapping + in-depth analysis), and a fully open validation infrastructure using Python and R scripts. Pilot testing on 24 studies achieved excellent inter-rater reliability (ICC = 0.92). All extraction templates, codebooks, and validation scripts are publicly released alongside the protocol to promote transparency, reproducibility, and cumulative science in the field of inclusive STEM technologies.

Abstract. Location-based recommender systems have achieved considerable sophistication, yet none provides a formal, lattice-theoretic representation of prerequisite dependencies among points of interest — the semantic reality that meaningfully experiencing certain locations presupposes contextual knowledge gained from others — nor the structural guarantees that such a representation entails. We introduce Exploration Space Theory (EST), a formal framework that transposes Knowledge Space Theory into location-based recommendation. We prove that the valid user exploration states — the order ideals of a surmise partial order on points of interest — form a finite distributive lattice and a well-graded learning space. These structural results yield four direct consequences: linear-time fringe computation, a validity certificate guaranteeing that every fringe-guided recommendation is a structurally sound next step, sub-path optimality for dynamic-programming path generation, and provably existing structural explanations for every recommendation. Building on these foundations, we specify the Exploration Space Recommender System (ESRS) — a memoized dynamic program over the exploration lattice, a Bayesian state estimator with beam approximation and EM parameter learning, an online feedback loop enforcing the downward-closure invariant, an incremental surmise-relation inference pipeline, and three cold-start strategies, the structural one being the only approach in the literature to provide a formal validity guarantee conditional on the correctness of the inferred surmise relation. All results are established through proof and illustrated on a fully traced five-POI numerical example.

Abstract. Learning dashboards aim to support decision-making throughout the learning process. However, their widespread adoption remains limited. While this observation can be attributed to their relatively recent emergence, research indicates that it is also due to a lack of stakeholder participation in the design process. To address this issue and involve users, we propose an integrated approach using a participatory design tool for the ideation phase and a generative tool for prototyping. We analyze these tools in terms of usability, support for participation, and user experience across the different phases. The analysis of feedback received demonstrates that these tools facilitate both end-user participation and developer user experience.

@misc{sadallah2023edarxiv,
  author = {Sadallah, Madjid and Gilliot, Jean-Marie},
  title = {Fostering Collaborative and Creative Design of Learning Dashboards: An Empowered Participatory Approach and Tools},
  year = {2023},
  howpublished = {EdArXiv},
  doi = {10.35542/osf.io/q78yn}
}

Abstract. The MOBILES RecSys system is an autonomous recommendation application designed to enhance the experience of international students using the MOBILES app. The app enables users to explore the city and enrich annotations based on their own experiences, offering a user-friendly interface for navigating urban routes and collecting geolocated data.

Abstract. The MOBILES project focuses on studying the spatial experiences and socialization of international students through their digital practices, combining sociolinguistics, geography, and computer science. The application allows international students to create rich geolocated annotations (text, photos, icons, moods) and share them.

@misc{lefevre2024mobiles,
  author = {Lefevre, Marie and Benedetto, Rapha\"{e}l and Sadallah, Madjid and Chevalier, Dominique and Cunty, Claire and Grassin, Jean-Fran\c{c}ois and Joliveau, Thierry and Mathian, H\'{e}l\`{e}ne and Scheffler, Camille},
  title = {MOBILES: A Mobile Application for Documenting the Urban and Social Experiences of International Students},
  year = {2024},
  note = {Software},
  howpublished = {\url{https://hal.science/hal-04771100}}
}

Abstract. Providing high-quality content is of utmost importance to drive successful reading. Besides, designing documents that are received the way the author wishes has always been difficult, and the digital world increases this difficulty by multiplying the possibilities related to mixed medias and interactivity. This compels authors to continuously review the delivered content to meet readers' needs. Yet it remains challenging for them to detect the comprehension barriers that may exist within their documents, and to identify how these latter can be improved accordingly. In this thesis, we focus on an educational context, where reading is a fundamental activity and the basis of many other learning activities. We propose a learning analytics approach for assisting course authors to maintain their courses to sustain learning. The proposals are based on theoretical background originated from research on learning analytics, reading comprehension and content revision. We advocate "usage-based document reengineering", a process defined as a kind of reengineering that changes document content and structures based on the analysis of readers' usages as recorded in their reading traces.

Abstract. The delivery of high-quality content is essential for effective reading comprehension and, by extension, successful learning. However, ensuring that educational materials are interpreted as intended by their authors remains a persistent challenge. This challenge is amplified in the digital age, where multimedia elements and interactivity add layers of complexity to content consumption. As a result, authors must continually revise their materials to meet the evolving needs of learners. Detecting comprehension barriers and identifying actionable improvements within documents is a complex task, particularly in the educational domain, where reading is a cornerstone of learning. This report presents an analytical approach to support course authors in the reengineering of their educational content, leveraging learning analytics and session-based indicators derived from learner activity logs.