Projects | Academic Page

Complex Networks Analysis

Thu, 31 Oct 2019 12:00:00 +0200

Complex Networks Analysis is the methodological core of the team’s work: the community detection, temporal analysis, and benchmarking methods developed here support the team’s more applied research directions.

On the methods side, the team develops and benchmarks community detection algorithms, anomaly and change point detection in temporal interactions, and comparative studies of detection methods. Ongoing PhD work extends this line to change point detection directly in linkstreams (continuous interaction sequences, as opposed to time-sliced snapshots).

Applied to the evolution of communities over time, this work covers the enumeration of temporal motifs describing community evolution, and the visualization of community evolution using Sankey diagrams. A related research direction investigates how large online collectives, such as Wikipedia or open-source software communities, grow and structure themselves over time, looking for invariant mechanisms behind their transitions between phases of low and high activity, and how individual contributors’ roles and trajectories relate to that collective-level structuring.

Community Dynamics & Individual Trajectories

Sat, 18 Jul 2026 12:00:00 +0200

This work analyzes and tracks how communities in temporal networks evolve, combining algorithm choice and benchmarking with several current and upcoming directions.

The team compares and benchmarks community detection algorithms for their ability to track how communities evolve, including a benchmarking framework built on customizable, realistic ground truths, and visualizes community evolution through purpose-built diagrams. These methods have been applied to real platforms, from crowdfunding communities to community dynamics and echo chambers on YouTube.

One current direction extends change point detection directly within linkstreams (continuous sequences of interactions, rather than time-sliced snapshots), to distinguish a lasting shift in a community from ordinary fluctuation without the sampling bias snapshots can introduce.

Another current direction looks at building domain-relevant metrics into community detection and clustering algorithms themselves, for example by weighting interactions or choosing filtering criteria that reflect what actually matters for a given application, explored on large real-world graphs such as Wikipedia’s.

Looking ahead, an upcoming PhD track will extend this work to individual trajectories: how members join, contribute, and move between communities over time, tracing each person’s evolving position within the community structures these methods detect.

Computational Social Science

Sat, 18 Jul 2026 12:00:00 +0200

This work applies network analysis methods to social phenomena, building on interdisciplinary collaborations bridging network science with media studies, economics, and sociology.

A recurring focus is echo chamber detection: characterizing community dynamics and echo chambers on platforms such as YouTube, drawing on analyses of very large comment datasets spanning millions of comments across tens of thousands of videos.

A related direction studies the massification of large-scale online collectives, such as Wikipedia or open-source software communities: how they grow, structure themselves, and sustain participation over time, an inherently interdisciplinary question bringing together sociology, economics, and management alongside network science.

The broader ambition is to understand team and social network interactions at scale, connecting network structure to questions of information pluralism and online public debate.

Pedagogical AI Agents

Sat, 18 Jul 2026 12:00:00 +0200

Educational chatbots built on large language models show real promise for learning outcomes, but most delegate pedagogical decisions such as what to teach and how to the model itself, making their tutoring strategies difficult to trace, evaluate, or reproduce, and leaving them prone to the error rates and domain variability that still make LLMs unreliable as autonomous instructors. This work instead formalizes the instructor’s own pedagogical reasoning into an explicit, inspectable layer: identifying what a student needs, selecting a didactic approach, and only then handing the wording over to a language model acting purely as a linguistic executor. The longer-term ambition is a Socratic companion that accompanies students throughout a course rather than just answering in isolation, treating an unclear question as an opportunity to guide the student toward a better one rather than as a failure to route, and keeping each interaction anchored in the instructor’s own teaching material and practice.

A first system, deployed for a university course on dimensional modeling, was evaluated on real student questions collected in production.

A second, more ambitious system extends this architecture with a library of discursive strategies and a multi-zone learner model, moving from a simple assistant towards a full Intelligent Tutoring System.

Learning Analytics

Wed, 05 Sep 2018 14:19:39 +0200

This work aims to use techniques from the analysis of social networks and machine learning to produce Learning Analytics (LA) that will support learners and teachers in changing their practices.

We are particularly interested in how the analysis of social interactions allows us to obtain insights into the dynamics of learning strategies.

This work is supported in particular by the post-doctoral work of Raphaël Charbey and the ongoing thesis work of Michael Kamau Wahiu.

Breast Cancer Risk Assessment

Wed, 05 Sep 2018 14:11:02 +0200

According to the World Health Organization, starting from 2010, cancer will become the leading cause of death worldwide. Prevention of major cancer localizations through a quantified assessment of risk factors is a major concern in order to decrease their impact in our society. In this project our objective was to find and to evaluate modeling methods easily readable by a physician.

This work is based on the thesis work of Gauthier Emilien (PhD defense in 2013).

Immersive Analytics

Fri, 30 Mar 2018 18:26:13 +0200

This work aims to develop new methods of data exploration and analysis using immersion techniques in 3D environments.

We are particularly interested here in the development of graph visualization techniques to allow:

an intuitive visualization of the centrality of nodes (in the sense defined in the field of social network analysis by closeness, betweenness, pagerank…)
the ability to navigate between several levels of detail (from the community graph to the communities themselves)

This work is based on the ongoing thesis work of Piriziwè Kobina.

Higher Education Pedagogy

Fri, 30 Mar 2018 18:04:08 +0200

The purpose of this work on pedagogy in higher education is to take a step back on the various pedagogical innovations implemented in my teaching activities at IMT Atlantique.

Several topics are covered:

student motivation and autonomy
skill assessment
peer review
the management of heterogeneous populations

Expert Knowledge Integration in Data Mining

Wed, 13 Dec 2006 12:00:00 +0200

This is the earliest research axis, developed during the PhD (2003-2006, under the supervision of Martine Collard) and the years immediately after.

The central idea was to integrate expert domain knowledge into the data mining process, rather than treating pattern extraction as a purely statistical problem. This led to the ExCIS methodology (Extraction using a Conceptual Information System), building a domain-specific ontology to guide dataset preparation and results interpretation, and to the IMAK interestingness measure, which evaluates extracted patterns by taking expert knowledge into account rather than relying on objective statistical measures alone.

Applications included transcriptome analysis (the KTA and HASAR frameworks, with atherosclerosis risk factor analysis) and gene expression data mining.