Cross-topic Rumour Detection in the Health Domain

Abstract

Nowadays information diffusion has become more and more immediate and fast thanks to social media and its services. However, lack of controls and moderation in resources as social microblogs often leads to spread unverified information, such as rumours, which can become a threat to the society. To improve life quality and good information diffusion, various automatic systems have been studied for rumour detection in microblogs at level of aggregation of posts, whereas a few effort has been tried to the most challenging scenario where the rumour has to be recognized at level of each single post. In this work, we direct our efforts towards individual post rumour detection: we investigate how features describing influence potential, personal interest and network characteristic perform on two different datasets of posts collected from Twitter using two different health-related keywords. As a further contribution, we study what happens in cross-topic tests, i.e. when the rumour detection system is trained with posts with an hashtag and tested on samples with a different one.