Posts by Collection

news

Our paper The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations has been accepted at ACL 2026!

April 07, 2026

I gave a talk at ISIR Young Scientists Day on our work on the GDN-CC dataset

May 26, 2026

Presented our work on the GDN-CC dataset representing the MLIA team during the ISIR Young Scientists Day. Check out the talk details.

I presented our work on The GDN-CC Dataset at ACL in San Diego, Calfifornia!

July 07, 2026

Our paper Give it Space! Explicit Disentangling of Positional and Semantic Representations in Encoders has been accepted at EMNLP 2026! see you there :)

August 21, 2026

We just published a new Preprint: Toward Collective-Centric Evaluation of Preference Inference.

September 02, 2026

Our new preprint Toward Collective-Centric Evaluation of Preference Inference for Participatory Democracy is now available on arXiv! We propose a collective-centric evaluation framework to analyze how preference inference models affect the collective opinion landscape in participatory democracy platforms. Check out the paper details.

portfolio

Portfolio item number 1

Short description of portfolio item number 1

Portfolio item number 2

Short description of portfolio item number 2

publications

Comment mesurer les biais politiques des grands modèles de langue multilingues?

Published in TALN (EALM workshop), 2025

Paper - Position paper on political bias in multilingual models.

Recommended citation: Lequeu, P. A., Labat, L., Cave, L., Lejeune, G., Yvon, F., & Piwowarski, B. (2026). The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations. arXiv preprint arXiv:2601.14944.

Fairbeat: Assessing and Mitigating Bias with the Composite Balance Score

Published in ECML-PKKD (Demo Track), 2025

Proceedings - A user interface for social bias evaluation in tabular datasets.

Recommended citation: Lequeu, P. A., Lagraa, S., Robin, G., & Ouedraogo, M. (2025, September). Fairbeat: Assessing and Mitigating Bias with the Composite Balance Score. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases (pp. 475-480). Cham: Springer Nature Switzerland.
Download Paper | Download Bibtex

The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations

Published in ACL, 2026

Paper - We introduce Corpus Clarification, a preprocessing framework of citizens consultation data which allow for ethical downstream analysis. We share a manually-annotated dataset based on the 2019 French consultation ‘Grand Débat National’, and a large automatically annotated dataset using SLMs finetuned for the task.

Recommended citation: Pierre-Antoine Lequeu, Léo Labat, Laurène Cave, Gaël Lejeune, François Yvon, and Benjamin Piwowarski. 2026. The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations. In Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 32976–33006, San Diego, California, United States. Association for Computational Linguistics.
Download Paper | Download Bibtex

Toward Collective-Centric Evaluation of Preference Inference for Participatory Democracy

Published in preprint, 2026

Preprint - Designed a new evaluation paradigm for preference inference and showed that recommender systems strongly distort the opinion landscape despite showing good results on standard metrics such as accuracy.

Recommended citation: Lequeu, P. A., Hafid, S., Lerner, P., Shafiabadi, N., Cave, L., Mas, D., ... & Yvon, F. (2026). Toward Collective-Centric Evaluation of Preference Inference for Participatory Democracy. arXiv preprint arXiv:2609.02990.
Download Paper

Give it Space! Explicit Disentangling of Positional and Semantic Representations in Encoders

Published in EMNLP, 2026

Preprint - explored how encoder-based models use absolute positional (AP) and relative positional (RP) information by explicitly disentangling positional and semantic representations. We find that the learned AP representations are low dimensional and used to encode document structure, while RP information is used as complementary to semantic matching.

Recommended citation: Lequeu, P. A., Barboule, C., & Piwowarski, B. (2026). Give it Space! Explicit Disentangling of Positional and Semantic Representations in Encoders. arXiv preprint arXiv:2605.30022.
Download Paper

talks

Grounding Synthesis in Human Votes: Improving Sparse Matrix Completion for Opinion Selection in Large-Scale Consultations

Mis à jour : May 15, 2026

Invited by the ParliView research team @ UC Dublin.

The GDN-CC dataset

Mis à jour : May 26, 2026

Presented our work “The GDN-CC Dataset: Automatic Corpus Clarification for AI-enhanced Democratic Citizen Consultations” representing the MLIA team for a day-long seminar on ISIR’s research.

Pierre-Antoine Lequeu