FrenchNews-7 is a newly presented benchmark for classifying French-language news articles by editorial desk across multiple French publishers. The dataset combines a large corpus of articles from various outlets with a seven-class taxonomy derived from URL patterns, allowing researchers to train and evaluate models in a cross-publisher setting.

The benchmark's authors fine-tune CamemBERT, a French-language transformer model, to perform the classification task. Because the taxonomy is derived from URLs and the corpus spans multiple outlets, the benchmark is intended to test how well models generalize to unseen publishers, rather than learning outlet-specific quirks.

The resource is likely to be useful for news aggregation, content recommendation, and media monitoring systems that need consistent section labels across heterogeneous sources. As a single-source announcement, the paper does not offer competing views; its contribution is the benchmark itself.