Fake-EmoReact 2021

Welcome to Fake-EmoReact 2021, the shared task of SocialNLP 2021 (in conjunction with NAACL 2021).

With the development of the Internet, misinformation gains a lot of attention in research fields and social issues. Benefit from the convenience of sharing information on online platforms, fake news could be spread rapidly from one to another, especially on social media. The misinformation could seriously influence modern life in many ways. The situation is even worse, due to the pandemic of Covid-19. Seeing the high impact of fake news on society, we eager to mitigate the effect of fake news by applying NLP techniques. For the reasons above, We design the Fake-EmoReact challenge.

We set the goal of this challenge to detect fake news tweets on Twitter as soon as possible from their text, replies, and most interestingly GIF responses. In this challenge, a total of 209,008 sequential utterances and their fake news labels were provided to the participants who registered for this shared task. Different from the conventional fake tweets detection, we especially focus on tweets with at least one GIF response. Each GIF response was labeled with the top 100 most popular categories of the animated GIFs, which provides emotion category. We have both the textual and visual data for participants to develop models for fake tweets detection. The challenge is defined as given an unlabeled tweet and its GIF response, the model should predict the label of tweet as fake or real. The positive samples are tweets with the hashtag #FakeNews where the Covid related ones are included, and the negative samples are from the previous SocialNLP challenge EmotionGIF (see https://sites.google.com/view/emotiongif-2020/ for details).

Important: Technical Papers for this task will not be archived