Abstract

Nowadays, users seek information from social media. Platforms such as Twitter, Facebook have become a major information source in people’s everyday life. Misinformation, however, pose negative effects on users, the platforms, corporates and governments. In this paper, we conduct a pilot study on a large scale dataset of false-proven rumors which spread on Weibo¹. We perform a through analysis on the dataset, including spreading patterns, credibility perceptions of the audiences, linguistic patterns as well as sentiment analysis. We demonstrate that rumors fall into very diverse categories, and the spreading patterns of rumors vary largely between content categories. People who construct the rumors also use more persuasive languages as well as negative sentiments, which could potentially help propagating the rumor. To our surprise, using the linguistic patterns and sentiment patterns alone, we could classify rumors from random public stream tweets with 93% of accuracy. To our best knowledge, this is the first large scale study on general rumors (without specifying domains or topics) in microblog. We provide detail analysis which could shed important light on following studies.

*The structure of this paper:
1) how we get the dataset?
2) is the dataset good enough to study?
3) why people are reporting rumors?
4) categories of rumors?
5) linguistic cues in rumors? and the association with spreading? (partially proved very very hard)
6) distinguish rumor tweets from random tweets, algorithmically