FRAME (Fact-Checking Reinforced by AI for Media outlets)
- Priority Area Boosting Fact-checking Activities in Europe
- Funding Round / Action Types 4th Funding Round – Scale-up Projects
- Year 2023, 2024
- Country France
- Project Status Completed
Fact-checking is about tracing and debunking disinformation campaigns, but doing so requires being able to monitor the use of certain keywords by public figures.
If there is a wide range of private actors for social listening, there is only one French actor that provides TV/radio listening (pluralisme.fr). What if we could combine social listening and TV/radio listening? It would certainly help with the tracing of disinformation campaigns in the public debate, without focusing solely on social media. This is FRAME, Fact-checking renforcé par Apprentissage automatique pour les médias.
Moreover, what if there was a tool that could match statements of public figures in video and audio, with everything a media outlet has previously fact-checked? This would be of great value: once a statement containing disinformation has been fact-checked, you could then monitor the reactions to the fact-checks: does the claim continue spreading? If so, who spreads it, where and on which platform? You could then reuse the original fact-checked content, albeit slightly modified for the present case, for fast and efficient fact-checking. It would also be useful for studies on the spread of disinformation and the impact of fact-checking. Moreover, the public could be given access to a public version and verify the spread of a statement.
FRAME is not automated fact-checking, but it would support fact-checkers’ daily work.
At first, FRAME aims at using machine-learning models using spoken language processing to help Les Surligneurs detect and trace law-related disinformation campaigns. At a later stage, other interested fact-checking organisations will be able to adapt the tool to their own publications.
The grant will fund access to a platform using state-of-the-art technology that aggregates automatic transcripts of audio and video sources containing the statements of public figures. Filters will be set up to search for statements according to different criteria.