Educational chatbots powered by large language models (LLMs) show promising effects on learning outcomes, yet most systems delegate pedagogical decisions such as content selection and didactic structuring implicitly to the LLM, making tutoring strategies difficult to trace, evaluate, and reproduce. This paper presents a didactical-driven teacher assistant for a French-language university course on dimensional modelling, operating without commercial LLM budget or GPU infrastructure. The architecture formalises the instructor's pedagogical reasoning into deterministic modules that handle intent detection, concept linking, and didactic approach selection before any text is generated; the LLM acts solely as a linguistic executor. Evaluation on 195 authentic student questions addresses two research questions. First, we show that standard semantic retrieval alone does not reliably recover the pedagogically required content, thereby justifying the upstream orchestration strategy adopted in our architecture (RQ1). Second, compared to free-tier LLMs whose detection performance varies widely across models and which produce errors silently, the deterministic pipeline achieves high pair precision (73%) with full traceability and explicit abstention, though its limited coverage confirms that the detection strategy requires further refinement (RQ2).

Graph models help understand network dynamics and evolution. Creating graphs with controlled topology and embedded partitions is a common strategy for evaluating community detection algorithms. However, existing benchmarks often overlook the need to track the evolution of communities in real-world networks. To address this, a new community-centered model is proposed to generate customizable evolving community structures where communities can grow, shrink, merge, split, appear or disappear. This benchmark also generates the underlying temporal network, where nodes can appear, disappear, or move between communities. The benchmark has been used to test three methods, measuring their performance in tracking nodes' cluster membership and detecting community evolution. Python libraries, drawing utilities, and validation metrics are provided to compare ground truth with algorithm results for detecting dynamic communities.

Given an arbitrary group of computers, how to identify abnormal changes in their communication pattern? How to assess if the absence of some communications is normal or due to a failure? How to distinguish local from global events when communication data are extremely sparse and volatile? Existing approaches for anomaly detection in interaction streams, focusing on edge, nodes or graphs, lack flexibility to monitor arbitrary communication topologies. Moreover, they rely on structural features that are not adapted to highly sparse settings. In this work, we introduce MAD, a novel Multi-scale Anomaly Detection algorithm that (i) allows to query for the normality/abnormality state of an arbitrary group of observed/non-observed communications at a given time; and (ii) handles the highly sparse and uncertain nature of interaction data through a scoring method that is based on a novel probabilistic and multi-scale analysis of sub-graphs. In particular, MAD is (a) flexible: it can assess if any time-stamped subgraph is anomalous, making edge, node and graph anomalies particular instances; (b) interpretable: its multi-scale analysis allows to characterize the scope and nature of the anomalies; (c) efficient: given historical data of length N and M observed/non-observed communications to analyze, MAD produces an anomaly score in O (NM); and (d) effective: it significantly outperforms state-of-the-art alternatives tailored for edge, node or graph anomalies.

In this paper, we present a new egocentric metaphor for graph visualization that consists in positioning a graph between two concentric spheres of different radii. It improves the expansion of nodes in space, contrary to 3D spatialization algorithms. The edge drawing is optimized by pushing all the edges into the area delimited by our two concentric spheres so that a user can move freely without being encumbered by edges. Our new metaphor also makes it possible to reduce the display angles in order to have a global view of the graph without leaving the egocentricity.

In this paper we propose improvements to the 3D radial layouts that make it possible to visualize centrality mea- sures of the nodes in a graph. Our improvements mainly relate edge drawing and the evaluation of the 3D radial layouts. First, we projected not only the nodes but also the edges onto the visualization surfaces in order to reduce the node overlap that could be observed in previous 3D radial layouts. Secondly, we proposed a human-centered evaluation in order to compare the efficiency score and the time to complete tasks of the 3D radial layouts to those of the 2D radial layouts. The evaluation tasks proposed are related to the central nodes, the peripheral nodes and the dense areas of a graph. The results showed that 3D layouts can perform significantly better than 2D layouts in terms of efficiency when tasks are related to the central and peripheral nodes, while the difference in time is not statistically significant between these various layouts. Additionally, we found that the participants preferred interacting with 3D layouts over 2D layouts.

L’informatique est une discipline exigeante à bien des niveaux, et son enseignement a suscité beaucoup de questionnements dans le domaine de la pédagogie. Quand s’ajoutent aux difficultés d’ordre disciplinaire, des éléments de complexité plus fort encore comme l’hétérogénéité d’un groupe d’étudiants (niveaux de compétences différents, formation et parcours professionnels différents, origines ethniques et culturelles différentes), les attendus des enseignants et des étudiants sont forcément mis à mal. Cet article propose l’évaluation d’un dispositif pédagogique par le jeu de rôle ayant pour but de prendre en compte les difficultés liées à l’altérité des étudiants et de permettre de les surmonter de manière ludique et motivante, en prenant en compte les spécificités de chacun au sein du groupe.

Engaging students in peer assessment is an innovative assessment process which has a positive impact on students learning experience. However, the adoption of peer assessment can be slow and uncomfortably experienced by students. Moreover, peer assessment can be prone to several biases. In this paper, we argue that the analysis of peer assessment interactions and phenomena can benefit from the social network analysis domain. We applied a graphlet-based method to a dataset collected during in-class courses integrating a peer assessment platform. This allowed for the interpretation of networking structures shaping the peer assessment interactions, leading for the description of consequent peer assessment roles and their temporal dynamics. Results showed that students develop a positive tendency towards adopting the peer assessment process, and engage gradually with well-balanced roles, even though, initially they choose mostly to be assessed by teachers and more likely by peers they know. This study contributes to research insights into peer assessment learning analytics, and motivates future work to scaffold peer learning in similar contexts.

This paper aims to understand student learning engagement in an online course. We describe an empirical study we conducted to investigate learner profiles when interacting with learning content. This study is based on data records about student online navigation and took place in the context of a hybrid environment. The obtained results showed that students mostly select assessment activities and visit the online course content without engaging deeply in the learning activities. This leads us to conclude on the role of assessment to motivate and engage students and on the importance of thinking out the design of the hybrid course. Finally, future work is motivated to study how to provide effective interactions with course content and how this can impact learning engagement and course design.

Discovering community structure in complex networks is a mature field since a tremendous number of community detection methods have been introduced in the literature. Nevertheless, it is still very challenging for practitioners to choose in each particular case the most suitable algorithm which would provide the richest insights into the structure of the social network they study. Through a case study of the French crowdfunding platform, Ulule, this paper demonstrates an original methodology for the selection of a relevant algorithm. For this purpose we, firstly, compare the partitions of 11 well-known algorithms. Then, bivariate map based on hub dominance and transitivity is used to identify the partitions which unveil communities with the most interesting size and internal topologies. These steps result in three community detection methods relevant for our data. Finally, we add the socioeconomic indicators, meaningful in the framework of the crowdfunding platform, in order to select the most significant algorithm of community detection, and to analyze the cooperation patterns among the platform's users and their impact on success of fundraising campaigns. In line with previous socioeconomic studies, we demonstrate that the social concept of homophily in online groups really matters. In addition, our approach puts in light that crowdfunding groups may benefit from diversity.

The authors seek to understand where and when YouTube comment sections are hit by brutal or aggressive debates and the factors that limit or foster this aggressive behaviour. To do so, they use a device at the crossroads of automated language processing and the political economy of media. The study is based on 2,209,206 comments, 1,184,859 of which are found in discussion threads, from the comment sections of 46,090 videos. The videos are taken from the YouTube channels of 57 different French media channels representing a variety of positions and institutional categories.

This paper presents new methods of 3D visualization of graphs that allow to highlight nodes structural centrality. These methods consist in projecting, along the vertical axis, 2D graph representations on three 3D surfaces: 1) a half-sphere; 2) a cone and 3) a torus portion. The transition to 3D allows to better handle the visualization of complex and large data that 2D techniques are generally unable to provide. The 3D radial layout techniques reduce nodes and edges overlap and improve, in some cases, the perception of nodes connectivity by exploiting differently or better the display space.

Our empiral study of the crowdfunding platform Ulule brings out the existence of a cross-projects social network at the platform scale. This social network has a catalyzing effect on the crowdfunding campaigns. The results highlight five sub-groups of contributors within the social network – sponsors, followers, precursors, specialists and collaborative specialists. Then, the analysis underlines a global and positive impact of the social network on the success rate. This positive influence becomes even more significant in the presence of sub-groups of contributors characterized by some thematic specialization and higher rates of collaboration.

This paper presents new methods of 3D visualization of graphs: projections on a half-sphere, a cone and a portion of torus. These methods allow to illustrate the importance of nodes in the graph in terms of centrality.

Comprendre la dynamique des interactions entre individus dans les reseaux sociaux est un enjeu important, notamment via la dynamique communautaire. Le resultat d'algorithmes de clustering temporel aboutit a la production de complexes graphes d'evolution de communautes que nous identifions a des graphes diriges acycliques aux aretes colorees et avec attribut sur les noeuds. Dans cet article, nous proposons une approche basee sur l'enumeration de motifs dans les graphes d'evolution de communautes, permettant de decrire le cycle de vie des communautes via la mise en evidence de differentes structures a l'echelle mesoscopique, c'est-a-dire a un niveau intermediaire entre celui des sommets et celui du graphe lui-meme. Nous proposons egalement une methode originale de coloration des aretes qui permet de reduire le nombre de motifs possibles et rendre ainsi plus aisee l'analyse de ces structures. Enfin, une experimentation basee sur un jeu de donnees d'echanges de commentaires sur YouTube permet d'illustrer la mise en oeuvre de notre approche.

L’analyse empirique de la plateforme de crowdfunding Ulule montre l’existence d’un réseau social transverse aux projets qui présente, à l’échelle de la plateforme, un effet catalyseur sur les campagnes de recherche de fonds. Les résultats mettent en évidence cinq sous-groupes de contributeurs au sein du réseau social de la plateforme–sponsors, suiveurs, précurseurs, spécialistes et spécialistes collaboratifs. Tandis qu’un impact positif global du réseau social sur le taux de succès des projets est mis en lumière, ce résultat est particulièrement significatif en présence de sous-groupes de contributeurs marqués par une spécialisation thématique et une dimension collaborative accrues

There is a growing interest in how data generated in learning platforms, especially the interaction data, can be used to improve teaching and learning. Social network analysis and machine learning methods take advantage of network topology to detect relational patterns and model interaction behaviors. Specifically, small induced subgraphs called graphlets, provide an efficient topological description of the way each node is embedded in the meso-scale structure of a network. Here we propose to detect the roles occupied by the different participants, students and teachers, in the successive phases of courses modeled by a sequence of static snapshots. The detected positions, obtained thanks to graphlet enumeration combined with a clustering method, reveal the different roles observed in each snapshot. We also track the role changes through the overall sequence of snapshots. We apply our method to the Sqily platform and describe the mutual skill validation process. The detected roles, the transitions between roles and a overall visualization through Sankey diagrams help interpreting the course dynamics. We found that some roles act like necessary steps to engage students within an active exchange process with their classmates.