Rumour Detection in Benchmark Dataset using Attention-Based Residual Networks

Abstract

With the meteoric advancements in social media, a variety of information has become readily accessible to the public. Social media has become paramount and it influences people significantly through information. The sheer volume of information diffusion has led to an imperative need for questioning the tangibility of information. Rumors are an imperious threat to the credibilityof the information sources. Rumors and non-rumors have to be meticulously separated from each other so that only the verified information reaches the public. This makes it necessary to look into the development of such tools or models that can detect rumors at an early stage and help in curbing their spread. In this paper, we have proffered anAttention-based Residual Network (ARN) model for rumour detection which employs residual blocks having skip connections, in combination with an attention mechanism. An early fusion strategy is used to combine the context-based (text + meta-features) and user-based features before feeding the combination to the ARN model, which outputs the final label as rumour or non-rumour. We have evaluated our proposed model on the PHEME dataset and the results validate superior classification performance to the state-of-the-art.