Detecting Misinformation and its Sources on Social Media

Over the last decade, social networks have become increasingly popular for news consumption due to their easy access, rapid dissemination, and low cost. However, social networks also allow for the wide spread of “fake news”, i.e., the spread either by accident, lack of proper knowledge or deliberately of news having false information. Fake news on social networks can have significant negative effects on society. Therefore, detecting fake news on social networks has recently become an emerging research area that is receiving considerable attention. There are several types of spreaders, and we focus on those who spread misinformation in social networks, and our research particularly focuses on those who spread it on Twitter during the Covid-19 health crisis. In this work we propose a solution that can be used by users to detect and filter sites having false and misleading information on the one hand, and by social network administrators to reduce the spread of false information on their platform on the other. We use simple and carefully selected characteristics of the title of the site where the information comes from and the content of the tweet to accurately identify false information. The approach used in this paper is to adopt a lightweight and efficient architecture for misinformation detection in the first step and a recommender-based architecture to determine which users are more likely to share the misinformation detected. Our Bidirectional GRU model performed 95.3% f1-score on the COVID-19 health misinformation dataset thus placing itself above the state of the art on this data set and our top-10 tweets not to recommend to a user model scored around 30%. code repository of this paper.

Recommended citation: Alex Kameni, 2022
Download Paper | Download Slides