The voice of silence: interpreting silence in truth discovery on social media

Abstract

This paper enhances the interpretation of silence for purposes of truth discovery on social media. Most solutions to fact-finding problems from social media data focus on what users explicitly post. Absence of a post, however, also plays a key role in interpreting veracity of information. In this paper, we focus on (absent links in) the retweet graph. A user might abstain from propagating content for many potential reasons. For example, they might not be aware of the original post; they might find the content uninteresting; or they might doubt content veracity and refrain from propagation (among other reasons). This paper formulates a joint fact-finding and silence interpretation problem, and shows that the joint formulation significantly improves our ability to distinguish true and false claims. An unsupervised algorithm, Joint Network Embedding and Maximum Likelihood (JNEML) framework, is developed to solve this problem. We show that the joint algorithm outperforms other unsupervised baselines significantly on truth discovery tasks on three empirical data sets collected using the Twitter API.