MSSF-GCN: Multi-scale Structural and Semantic Information Fusion Graph Convolutional Network for Controversy Detection

Abstract

Detecting controversial posts on the web and social media play an important role in judging the authenticity of web information, measuring the influence of news and alleviating the polarized views. The controversy detection task has attracted widespread attention from researchers in the fields of computer science and social humanities sciences. However, previous works do not achieve: 1) preserve the reply-structure relationship with sentiment information; 2) integrate multi-scale structure and semantic information and provide interpretable results; 3) learn effectively topics and comments information related to the target post. To overcome the first limitation, we construct a Topic-Post-Comment-Sentiment Graph (TPCS Graph) for preserving the reply-structure and incorporate the sentiment information. For the second and third limitation, we propose Multi-scale Structural and Semantic Information Fusion Graph Convolutional Network (MSSF-GCN) for post-level controversy detection. Moreover, we build a multilingual dataset for controversy detection. We conduct comprehensive experiments on two real-world datasets and the results show that the proposed method exhibits comparable or even superior performance.