Fake News Detection with Heterogenous Deep Graph Convolutional Network

Zhezhou Kang, Yanan Cao, Yanmin Shang, Tao Liang, Hengzhu Tang, Lingling Tong

Abstract: Fake news detection is a challenging problem due to its tremendous real-world political and social impacts. Previous works judged the authenticity of news mainly based on the content of a single news, which is generally not effective because the fake news is often written to mislead users by mimicking the true news. This paper innovatively utilizes the connection between multiple news, such as their relevance in time, content, topic and source, to detect fake news. We construct a heterogeneous graph with different types of nodes and edges, which is named as News Detection Graph (NDG), to integrate various information of multiple news. In order to learn deep representation of news nodes, we propose a Heterogenous Deep Convolutional Network (HDGCN) which utilizes a wider receptive field, a neighbor sampling strategy and a hierarchical attention mechanism. Extensive experiments carried on two real-world datasets demonstrated the effectiveness of our work in solving the fake news detection problem.