French-TV-Headline-Classification is an encoder model specialized for the classification of French news headlines.
The model incorporate two different sets of categories: thematics and non-thematics. Headlines fill multiple functions in the informational design of television and not all of them are straight about actual topics.
Thematics categories
This covers all headlines about a classifiable event. The classification relies not only on the actual content but also on very distinctive manneer of phrasing the news, as the headlines are likely issues by different news service.
Typically crime news ("fait divers") uses a more daily/informal language than political news.
The available thematic categories are:
- Politique
- Relations internationales
- Santé
- Santé
- Sport
- Économie
- Éducation
- Culture
- Fait divers
- Débat
Non-thematic categories
This includes headlines that are not directly describing an event but filling multiple meta-communication tasks helping to better contextualize the news broadcasts.
The available non-thematic categories are:
- Attribution (typically caption for the journalists that made a news reports)
- Clip (a musical clip with credits)
- Citation (a quote attributed to a specific person)
- Intervenant (the name/bio of a person speaking)
- Sondage (results of a poll)
- Méta (any kind of meta indication related to the broadcast)
- Inclassable (can't be classified under any of the previous categories)
Examples
- Dune 2 : Villeneuve dévoile les coulisses du film => Culture
- Nucléaire/espace : Poutine veut s'y implanter => Relations internationales
- Philippe Chalmin, économiste, spécialiste des marchés de matières premières => Intervenant
- Belloubet : "Faire de l'école un sanctuaire" => Citation
- Downloads last month
- 79