DECOR: Degree-Corrected Social Graph Refinement for Fake News Detection

Jiaying Wu,Bryan Hooi

DOI: https://doi.org/10.1145/3580305.3599298

2023-07-01

Abstract:Recent efforts in fake news detection have witnessed a surge of interest in using graph neural networks (GNNs) to exploit rich social context. Existing studies generally leverage fixed graph structures, assuming that the graphs accurately represent the related social engagements. However, edge noise remains a critical challenge in real-world graphs, as training on suboptimal structures can severely limit the expressiveness of GNNs. Despite initial efforts in graph structure learning (GSL), prior works often leverage node features to update edge weights, resulting in heavy computational costs that hinder the methods' applicability to large-scale social graphs. In this work, we approach the fake news detection problem with a novel aspect of social graph refinement. We find that the degrees of news article nodes exhibit distinctive patterns, which are indicative of news veracity. Guided by this, we propose DECOR, a novel application of Degree-Corrected Stochastic Blockmodels to the fake news detection problem. Specifically, we encapsulate our empirical observations into a lightweight social graph refinement component that iteratively updates the edge weights via a learnable degree correction mask, which allows for joint optimization with a GNN-based detector. Extensive experiments on two real-world benchmarks validate the effectiveness and efficiency of DECOR.

Social and Information Networks

What problem does this paper attempt to address?

### Problems the Paper Aims to Solve

This paper aims to address a key challenge in fake news detection, namely the impact of noisy edges in social graphs on Graph Neural Networks (GNNs). Specifically, the authors propose a new method—**Degree-Corrected Social Graph Refinement (DECOR)**—to improve the quality of social graphs in the task of fake news detection.

#### Main Issues
1. **Impact of Noisy Edges**: Existing studies typically use a fixed social graph structure, but these graphs contain noisy edges that severely affect the expressive power of GNNs.
2. **Limitations of Existing Methods**: Although existing Graph Structure Learning (GSL) methods can mitigate the problem of noisy edges, they usually rely on node features to update edge weights, which is computationally expensive and difficult to apply in large-scale social graphs.

#### Specific Goals
- **Identify Noisy Edges**: Identify which edges might be noisy by observing the degree distribution and co-participation patterns of news article nodes.
- **Downweight Noisy Edges**: Improve the accuracy of fake news detection by learning a degree-corrected social graph optimization framework (DECOR) to downweight noisy edges.
- **Joint Optimization**: Combine DECOR with GNNs for joint optimization to predict the authenticity labels of news.

### Research Contributions
1. **Empirical Findings**: Propose two new findings that degree and co-participation patterns are closely related to news authenticity.
2. **Theoretical Foundation**: Based on the Degree-Corrected Stochastic Blockmodel (DCSBM), propose DECOR to reduce noisy edges.
3. **Efficiency**: Compared to existing GSL methods, DECOR avoids high-dimensional feature input and has linear time complexity, increasing speed by 7.6 to 34.1 times.
4. **Effectiveness**: DECOR significantly improves the F1 score on two real-world datasets and consistently enhances performance even in label-scarce scenarios.

Through the above work, DECOR effectively addresses the issue of noisy edges in fake news detection, improving both the accuracy and efficiency of detection.