Rumor Detection on Twitter with Hierarchical Attention Neural Networks

Social media has become an integral part of peoples agenda life, however social media also plays a vital role in communication and information dissemination, making it an ideal platform for spreading rumors. Automatically debunking rumors at their stage of diffusion is known as early rumor detection. However, automatic rumor detection is technically very difficult and uncertain. In this work, we try to build a model to learn extracted features from tweets content by concentrating their posts and generate more meaningful representations to identify different type of rumors. This paper presents a deep hierarchical attention model on the basis of recurrent neural networks (RNN) for rumor representation learning and classification. Extensive experiments on real datasets which collected from social media web sites (mainly on twitter) demonstrate that the deep hierarchical attention based RNN model outperforms state-of-the-arts that rely on hand-crafted features.