Automatic Controversy Detection Based on Heterogeneous Signed Attributed Network and Deep Dual-Layer Self-Supervised Community Analysis
Year: 2025
Type: article
Abstract: Controversial analysis entails identifying and scrutinizing contentious events or issues within a specific domain. Detecting controversy, particularly on social media, poses challenges due to the need to consider numerous contents and user interactions. Unlike previous work, rather than studying controversy on limited knowledge of the topic discussion, we introduce multidimensional and heterogeneous information to conduct a comprehensive evaluation. Our method to quantify controversy is based on heterogeneous signed attributed network. This framework constructs a comprehensive heterogeneous signed attributed network to integrate diverse semantic and structural information on social media, encompassing user, retweet, comment, summary, topic, sentiment information, and user attributes. We introduce a deep dual-layer self-supervised algorithm for community detection and controversial analysis within this network. We gauge controversy levels in hot events via community distinctions, betweenness centrality, and user representations. Extensive comparisons of state-of-the-art baseline approaches are performed. We find that existing methods shows instability on different datasets. Whereas our approach integrates the multidimensional features of topics and also takes into account various computational results, our method consistently yields stable results. A user-friendly web server has been developed for accessible analysis of other hot events (http://www.csbg-jlu.site/contro/).
Source: Entropy
DOI: 10.3390/e27050473
Authors: Ying Li, Xiao Zhang, Yu Liang, Qianqian Li
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,137
Awards: G3397295082, G3466901669
OpenAlex ID: W4409926850