Heterogeneous Subgraph Transformer for Fake News Detection

Yuchen Zhang, Xiaoxiao Ma, Jia Wu, Jian Yang, Hao Fan

Keywords: fake news detection, misinformation and disinformation, subgraph mining, heterogeneous graph

Abstract:

Fake news is pervasive on social media, inflicting substantial harm on public discourse and societal well-being.
We investigate the explicit structural information and textual features of news pieces by constructing a heterogeneous graph with regard to the relations among news topics, entities, and content.
Through our study, we reveal that fake news can be effectively detected in terms of the atypical heterogeneous subgraphs centered on them.
These subgraphs encapsulate the essential semantics of news articles as well as the intricate relations between different news articles, topics, and entities. However, suffering from the heterogeneity of topics, entities, and news content, exploring such heterogeneous subgraphs remains an open problem.
To bridge the gap, this work proposes a hierarchical framework - heterogeneous subgraph transformer (HeteroSGT) - to exploit subgraphs in our constructed heterogeneous graph.
In HeteroSGT, we first apply a pre-trained dual-attention language model to derive textual features in accordance with word-level and sentence-level semantics.
Then, we employ random walk with restart (RWR) to extract subgraphs centered on each news. The extracted subgraphs are further fed to our proposed subgraph Transformer to encode the subgraph surrounding each news piece for quantifying its authenticity.
Extensive experiments on five real-world datasets demonstrate the superior performance of HeteroSGT over five baselines.
Further case and ablation studies validate our motivation in investigating the subgraphs centered on news and demonstrate that performance improvement stems from our specially designed components. The source code of HeteroSGT is available at https://github.com/HeteroSGT/HeteroSGT}

Track: Responsible Web

Submission Guidelines Scope: Yes

Submission Guidelines Blind: Yes

Submission Guidelines Format: Yes

Submission Guidelines Limit: Yes

Submission Guidelines Authorship: Yes

Student Author: Yes

Submission Number: 2204