Neighborhood-Order Learning Graph Attention Network for Fake News Detection

TL;DR

The paper addresses fake news detection on graph-structured data, noting that conventional GNNs limit information flow to fixed $L$-hop neighborhoods and thus miss distant yet informative signals. It introduces NOL-GAT, a dual-network architecture with Hop Network $\Phi$ and Embedding Network $\Psi$ that learns a per-node, per-layer neighborhood order via a differentiable $Gumbel$-Softmax over $\Gamma = \{0,\dots, d_g\}$, enabling adaptive and targeted message passing. This approach mitigates over-smoothing and over-squashing, reduces computational burden, and yields strong improvements in accuracy and macro-F1, especially under low-label settings, across five datasets. The results validate adaptive neighborhood-order learning as a practical and effective strategy for semi-supervised graph-based fake news detection.

Abstract

Fake news detection is a significant challenge in the digital age, which has become increasingly important with the proliferation of social media and online communication networks. Graph Neural Networks (GNN)-based methods have shown high potential in analyzing graph-structured data for this problem. However, a major limitation in conventional GNN architectures is their inability to effectively utilize information from neighbors beyond the network's layer depth, which can reduce the model's accuracy and effectiveness. In this paper, we propose a novel model called Neighborhood-Order Learning Graph Attention Network (NOL-GAT) for fake news detection. This model allows each node in each layer to independently learn its optimal neighborhood order. By doing so, the model can purposefully and efficiently extract critical information from distant neighbors. The NOL-GAT architecture consists of two main components: a Hop Network that determines the optimal neighborhood order and an Embedding Network that updates node embeddings using these optimal neighborhoods. To evaluate the model's performance, experiments are conducted on various fake news datasets. Results demonstrate that NOL-GAT significantly outperforms baseline models in metrics such as accuracy and F1-score, particularly in scenarios with limited labeled data. Features such as mitigating the over-squashing problem, improving information flow, and reducing computational complexity further highlight the advantages of the proposed model.

Figures (5)

• Figure 1: An overview of the proposed model NOL-GAT
• Figure 2: An illustration of the neighborhood limitation in standard GNNs. The graph consists of 7 nodes ($a$ to $g$) arranged in a simple path, where each node is connected to its adjacent neighbors. In a 2-layer GNN, node $a$ can only aggregate information from nodes $b$ and $c$, while information from more distant but potentially important nodes remains inaccessible.
• Figure 3: Macro-F1 (a) and accuracy (b) comparision of NOL-GAT and standard GATv2.
• Figure 4: Macro-F1 (a) and accuracy (b) comparision, for different amounts of labeled data
• Figure 5: Parameter sensitivity of k (number of nearest neighbors in $G_{KNN}$). (a): Fake.Br, (b): Fact-checked News, (c): FNN, (d): FakeNewsDetection, (e): FakeNewsData.