MSynFD: Multi-hop Syntax aware Fake News Detection

TL;DR

MSynFD tackles fake news detection by integrating multi-hop syntactic dependencies with a sequential Transformer, addressing syntax–semantics mismatch and prior word biases. It introduces a Subgraph Aggregation Attention mechanism over $d$-hop graphs, a sequential relative position-aware Transformer for semantic cues, and a keywords debiasing module, with a fusion-based detector. Empirical results on Weibo and GossipCop show state-of-the-art performance across key metrics, supported by comprehensive ablations and qualitative analyses. The approach enhances word perception scope, reduces bias from keywords, and demonstrates robustness when modalities or context are limited, with potential applicability to other fine-grained semantic tasks.

Abstract

The proliferation of social media platforms has fueled the rapid dissemination of fake news, posing threats to our real-life society. Existing methods use multimodal data or contextual information to enhance the detection of fake news by analyzing news content and/or its social context. However, these methods often overlook essential textual news content (articles) and heavily rely on sequential modeling and global attention to extract semantic information. These existing methods fail to handle the complex, subtle twists in news articles, such as syntax-semantics mismatches and prior biases, leading to lower performance and potential failure when modalities or social context are missing. To bridge these significant gaps, we propose a novel multi-hop syntax aware fake news detection (MSynFD) method, which incorporates complementary syntax information to deal with subtle twists in fake news. Specifically, we introduce a syntactical dependency graph and design a multi-hop subgraph aggregation mechanism to capture multi-hop syntax. It extends the effect of word perception, leading to effective noise filtering and adjacent relation enhancement. Subsequently, a sequential relative position-aware Transformer is designed to capture the sequential information, together with an elaborate keyword debiasing module to mitigate the prior bias. Extensive experimental results on two public benchmark datasets verify the effectiveness and superior performance of our proposed MSynFD over state-of-the-art detection models.

Figures (6)

• Figure 1: (a) A fake news example with misleading information is highlighted in yellow. The word correlations above show how irrelevant words affect the understanding of the center word 'our', then mislead the detection result; (b) A true news example including keywords marked in grey and words leading to potential prior bias list below. The left region of both (a) and (b) shows syntax-associated words towards the center word 'our' at the 3-hops case and the local structure of the syntactic dependency tree.
• Figure 2: Overview of our MSynFD fake news detection method.
• Figure 3: Comparison illustration of information propagation among (a) attention-based methods, (b) traditional GNN-based methods, and (c) our Multi-hop Syntax aware module.
• Figure 4: (a) Performance of the MSynFD model under different values of the parameter hops; (b) Performance of the MSynFD model and MSynFD ¬ KD under different values of the parameter max length.
• Figure 5: Case Study: Two pairs of cases from the Weibo and the GossipCop datasets, respectively. The center words focused by us are boldfaced, while the darker cell color indicates higher attention value, the yellow areas, orange areas, and red areas indicate the focus from the Semantic Aware Module, the SAA Module, or both focus.