Rumor Detection by Propagation Embedding Based on Graph Convolutional Network

Abstract

Detecting rumors is an important task in preventing the dissemination of false knowledge within social networks. When a post is propagated in a social network, it typically contains four types of information: i) social interactions, ii) time of publishing, iii) content, and iv) propagation structure. Nonetheless, these information have not been exploited and combined efficiently to distinguish rumors in previous studies. In this research, we propose to detect a rumor post by identifying characteristics based on its propagation patterns and other kinds of information. For the propagation pattern, we suggest using a graph structure to model how a post propagates in social networks, allowing useful knowledge to be derived about a post's pattern of propagation. We then propose a propagation graph embedding method based on a graph convolutional network to learn an embedding vector, representing the propagation pattern and other features of posts in a propagation process. Finally, we classify the learned embedding vectors to different types of rumors by applying a fully connected neural network. Experimental results illustrate that our approach reduces the error of detection by approximately 10% compared with state-of-the-art models. This enhancement proves that the proposed model is efficient on extracting and integrating useful features for discriminating the propagation patterns. (C) 2021 The Authors. Published by Atlantis Press B.V.