We are proud to share the first research publication from Le French News Lab: Beyond Hate Speech and Media Bias: Detecting Prejudicial Language in French Political News.
The paper introduces a theory-grounded task and dataset for identifying prejudicial identity framing in French political news — a phenomenon that sits between traditional hate speech categories and broader notions of media bias. It also presents benchmark results showing that fine-tuned transformer models substantially outperform lexical baselines on the task.
This publication marks an important step in our broader research agenda on French political discourse, media framing, and computational social science. It establishes a first research cornerstone for future work at Le French News Lab on harmful framing, public-language analysis, and French-language media narratives.
Results at a glance
0.820
F1 — CamemBERT-large
0.892
PR-AUC — CamemBERT-large
3
Models benchmarked
Read the full paper, view the benchmark results, and find the citation on the publication page.