Fact-checking consists of systematically verifying statements made by political leaders or elements of public debate, especially during election periods. Radio France has entered into a partnership with Polytechnique and Inria to develop artificial intelligence in its newsrooms. It intends to use it for the 2022 election campaign and is recruiting to complete the CEDAR team.

Fact-checking

This mode of journalistic processing, born in the United States, is a process of internal verification of news organizations, some employees have the function of verifying the accuracy of facts, figures or quotes reported by field journalists. Time magazine hired the first fact-checkers in 1923, but it was the development of the Internet that reinvented it in the 2000s: the verifications, made after the fact and no longer before, concern the elements of political speeches and more broadly of the public debate, especially during elections. Sites such as Factcheck.org or Politfact have been created to clarify the public debate by verifying and correcting misleading or confusing assertions. Newspapers are not to be outdone: Libération has launched its "Désintox" section and Le Monde, the blog "Les Décodeurs". Television and radio are also involved in fact-checking.

In her "Strategic Project 2018 -2023 for Radio France", Sybile Veil, CEO of the Radio France group since the end of last December, declared:

"The franceinfo label must be the guarantee for the public of "100% verified" information. The networking of fact-checking skills and the rise of Agence france info will make Radio France the hub of the news verification function, a crucial mission that must permeate all public broadcasting newsrooms."

The Inria - Ecole Polytechnique - RadioFrance partnership

The collaborative project between Inria, Ecole Polytechnique and Radio France to automatically detect infomercials widely reported in the media or on social networks focuses on two areas:
  • Accessibility of reference data.
Software tools for querying quality statistical sources (in particular INSEE, but also OECD, etc.) will be implemented in order to identify very quickly and accurately the reference data to verify a statistical statement. The aim is to build a complete system, starting with the acquisition and indexing of open reference data, passing through specialized information search algorithms, and completed by modules for the automatic detection of statistical assertions to be verified in written content. This axis will build on previous work of the team.
  • Creation of a body of arguments (evidence) related to controversial topics.
We need to identify reliable sources, such as research or popular science articles for each topic of interest and then create a corpus of structured arguments, based on the data sources. Each argument will keep the link to the original source, so that a journalist will be able to access the full context of the publication of a certain information. The corpus could be easily queried by the techniques proposed in axis 1.

In order for the project to become operational for the presidential elections, the partners are recruiting. They are urgently looking for :

  • One or two engineers with very good expertise in Python programming, collaborative and multi-version development (Git), and ideally knowledge of: web technologies, artificial intelligence (AI), especially natural language processing (NLP). Training on these topics can be provided by the team, but previous exposure to them will be a plus.
  • One or more student interns (M1 level or higher) with a solid foundation in Python and algorithms. Knowledge in AI, NLP, databases and web technologies will be appreciated. The project is broken down into multiple well-defined tasks requiring different skills; the work of each step will be part of a specific task.
The work will take place in the CEDAR team (Inria and LIX, CNRS and Ecole Polytechnique). It will be supervised by Ioana Manolescu (DR Inria), head of the CEDAR team, and by Oana Balalau (Starting Faculty Inria). Since 2013, the team has been developing a recognized expertise in content management techniques (data, text, etc.) for data journalism and journalistic verification, in particular in the ANR ContentCheck project on which this project is based, then in the framework of the IA SourcesSay chair. More recently, the team has obtained new results in online discourse analysis and argument extraction.

In a previous interview, Sybile Veil said:

"We are living in an information war today. While the multiplication of sources goes hand in hand with the impoverishment of the quality of information and the development of misinformation, we must preserve the ability to invest time and resources in quality information."

Translated from INRIA, l'Ecole Polytechnique et Radio France se lancent dans le fact-checking