French Tweet Corpus for Automatic Stance Detection

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6317–6322 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6317 French Tweet Corpus for Automatic Stance Detection Marc Evrard, R´emi Uro, Nicolas Herv´e, and B´eatrice Mazoyer Institut national de l’audiovisuel (INA) 18, avenue des Fr`eres-Lumi`ere 94366 Bry-sur-Marne Cedex {mevrard, ruro, nherve, bmazoyer}@ina.fr Abstract The automatic stance detection task consists in determining the attitude expressed in a text toward a target (text, claim, or entity). This is a typical intermediate task for the fake news detection or analysis, which is a considerably widespread and a particularly difficult issue to overcome. This work aims at the creation of a human-annotated corpus for the automatic stance detection of tweets written in French. It exploits a corpus of tweets collected during July and August 2018. To the best of our knowledge, this is the first freely available stance annotated tweet corpus in the French language. The four classes broadly adopted by the community were chosen for the annotation: support, deny, query, and comment with the addition of the ignore class. This paper presents the corpus along with the tools used to build it, its construction, an analysis of the inter-rater reliability, as well as the challenges and questions that were raised during the building process. Keywords: Stance Detection, French, Dataset, Fake News, Twitter, Inter-Rater Reliability, Natural Language Processing 1. Introduction Fake news proliferation is a conspicuous issue, which seems particularly difficult to overcome. Their detection is a complex task, requiring both a large and precise knowl- edge of the world. Indeed, most domains seem to be af- fected by fake news, while some rumors refer to very spe- cific topics that are known by only small groups of initiated people, which implies that extensive knowledge of all ex- isting concepts would be necessary to tackle this problem. Since a direct approach leading to a high accuracy does not yet seem practical, we are instead currently focusing on the gathering of information to describe and analyze fake news. One of the parameters we wish to investigate is the deter- mination of the stance of a text toward another text it relates to. This is a typical intermediate task of the fake news anal- ysis or detection, as the discussions concerning them are of- ten controversial and debated, especially on social networks (Zubiaga et al., 2018). Discussion threads about rumors are thus expected to exhibit contrasting stances. Moreover, it has been shown that the veracity of a rumor is correlated to the stance of discussion toward it (Ferreira and Vlachos, 2016; Enayet and El-Beltagy, 2017). We decided to resort to social networks to gather the text corpus since it is a major source of fake news. Accord- ing to the Reuters Institute (Nic et al., 2018) in their Digi- tal News Report of 2018, 36% of French adults use social media as a source of news. Although this share has de- clined since 2017, it is mostly due …

Content truncated to 500 words