Detection of Context-Varying Rumors on Twitter through Deep Learning

Abstract

The rapid exchange of information and large user base of online social networks such as Twitter or Facebook make them an ideal platform to gather the latest information. Peoples belonging to different parts of the world can easily share thoughts or updates on real-time events just by having an internet connected device. This low cost of information exchange and inadequacy of techniques which can check the veracity of shared information, gives birth to deliberate and the accidental spread of rumors i.e. pieces of information having uncertain truth at the time of posting. There exist techniques which detect rumors on online social networks by extracting patterns from pre-identified rumors, but these techniques are not sufficient to detect fast paced rumors (i.e. breaking news based rumors). Existing techniques require periodic update of rumor detecting patterns for identifying newly emerging rumors. In this paper, a deep learning model is proposed which required no periodic update of rumor related patterns to detect the rumorous information. The results clearly reveal how our approach outperformed state of the art methods of rumor detection.