We are proud to share the first research publication from Le French News Lab by Amr Sobhy: FrenchNews-7 — a cross-publisher French news classification benchmark.
FrenchNews-7 organizes 87,637 French news articles from 13 publishers into 7 shared editorial desk categories. The benchmark harmonizes publisher-specific editorial structures into a common taxonomy, making it possible to study cross-publisher coverage patterns and benchmark French desk classification systems.
The project is released as a small research ecosystem: a paper describing the benchmark design and evaluation, a public CamemBERT-base classifier on Hugging Face, and a manifest-only dataset artifact that supports reproducibility without redistributing publisher article text.
Results at a glance
0.847
Macro F1
87,637
Articles labeled
13
Publishers
7
Desk categories
Read the full paper, view the benchmark results, and find the citation on the publication page.