No, That Never Happened!! Investigating Rumors on Twitter

Abstract

We focus on two problems related to rumor detection. First, stance classification with respect to a rumor and, second, rumor veracity prediction. In the stance classification task, we aim to identify the users' stance toward the underlying rumor in a Twitter conversational thread, whereas, in the second problem, i.e., veracity prediction, we aim to verify the authenticity of a rumorous tweet (i.e., source tweet) in a conversational thread. We propose an MLP-based feature-driven model for veracity prediction and a hierarchical LSTM-based approach for detecting stances toward a rumor in a conversation thread. Evaluations show that our proposed system attained better performance in comparison with the various state-of-the-art systems on both the tasks.