Psychology-informed Rumor and Stance Detection

In recent years, online social networks have developed rapidly, the spread of rumors becomes convenient and rapid, so it is necessary to study rumor detection and stance classification. In social networks, users with different personality behaves differently during the spread of rumors and their stance is usually affected by their personality. Therefore, it is particularly important to analyze the personality of social network users. In this paper, we add the Big Five personality analysis to the model and design a multi-task learning model that combines rumor detection, stance classification, and personality recognition tasks. Firstly, daily tweets of users were crawled on the official Twitter website, and the Linguistic Inquiry and Word Count (LIWC) was used to obtain the true value of each user’s Big Five personality. Secondly, the shared Long Short-Term Memory (LSTM) layer is used to train the input embedding of rumor detection and stance classification tasks, and the Multilayer Perceptron (MLP) model is used to obtain the Big Five personality prediction value of each user. Finally, tweet embedding is combined with user personality embedding to jointly train the rumor detection and stance classification tasks. Compared with baseline rumor detection and stance classification models on the RumourEval and PHEME datasets, our model has the best performance on each task.