A writing style-based multi-task model with the hierarchical attention for rumor detection.

With the development of the Internet and social media, the harm caused by rumors has become more and more serious. Existing rumor detection methods focus on determining rumors by capturing their unusual textual content or communication structure, but fewer methods focus on the writing style of rumors. In order to identify rumors more effectively, we design and implement a multi-task rumor detection model with the hierarchical attention mechanism based on writing styles inspired by multi-task learning in this paper. The model combines a content-based rumor detection task and a writing style-based rumor detection task in a multi-task format, so that the two tasks can enhance their respective detection effects by interacting with each other during the model training process. In addition, we also use the hierarchical attention mechanism consisting of a word attention mechanism and a sentence attention mechanism to focus on words and posts that are more useful for rumor detection, which can reduce the interference of noise and further improve the detection accuracy. The experimental results of our model on the publicly available English Pheme dataset and Chinese Weibo dataset show that our model outperforms most of the existing better rumor detection methods. [ABSTRACT FROM AUTHOR]