Rumor Detection on Social Media with Bi-Directional Graph Convolutional Networks

Abstract: Social media has been developing rapidly in public due to its nature of spreading new information, which leads to rumors being circulated. Meanwhile, detecting rumors from such massive information in social media is becoming an arduous challenge. Therefore, some deep learning methods are applied to discover rumors through the way they spread, such as Recursive Neural Network (RvNN) and so on. However, these deep learning methods only take into account the patterns of deep propagation but ignore the structures of wide dispersion in rumor detection. Actually, propagation and dispersion are two crucial characteristics of rumors. In this paper, we propose a novel bi-directional graph model, named Bi-Directional Graph Convolutional Networks (Bi-GCN), to explore both characteristics by operating on both top-down and bottom-up propagation of rumors. It leverages a GCN with a top-down directed graph of rumor spreading to learn the patterns of rumor propagation, and a GCN with an opposite directed graph of rumor diffusion to capture the structures of rumor dispersion. Moreover, the information from the source post is involved in each layer of GCN to enhance the influences from the roots of rumors. Encouraging empirical results on several benchmarks confirm the superiority of the proposed method over the state-of-the-art approaches.

• The paper proposes a dual-layer Bi-GCN model that combines top-down and bottom-up graph convolution to capture rumor propagation and dispersion.
• The model incorporates root node feature enhancement to preserve influential source information across convolutional layers.
• In experiments, Bi-GCN outperformed traditional and deep learning methods, achieving 96.1% accuracy on the Weibo dataset.

Rumor Detection on Social Media with Bi-Directional Graph Convolutional Networks

This paper presents a novel approach to rumor detection on social media leveraging the structure of Bi-Directional Graph Convolutional Networks (Bi-GCN). Bi-GCN is designed to capture both propagation and dispersion characteristics of rumor spread by modeling social media interactions as graph structures.

Methodology and Model Design

The proposed model employs a dual-layered approach leveraging both Top-Down Graph Convolutional Networks (TD-GCN) and Bottom-Up Graph Convolutional Networks (BU-GCN). TD-GCN captures the propagation characteristics of rumors, whereas BU-GCN focuses on the dispersion nature across social media platforms. By integrating these two graph-based perspectives, Bi-GCN gains a comprehensive understanding of rumor diffusion.

The model enhances rumor detection by incorporating features from the source post at each convolutional layer. This root node feature enhancement ensures that the influential information from the initial point of rumor origination is adequately represented throughout the network layers.

Experimental Setup and Results

The model's effectiveness was evaluated on three datasets: Weibo, Twitter15, and Twitter16. Comparative analyses with existing models, including methods based on Decision Trees and SVMs with handcrafted features, demonstrated Bi-GCN's superior performance in detecting rumors. Deep learning models such as RvNN and PPC_RNN+CNN were also outperformed by Bi-GCN.

Key metrics for evaluation included accuracy and F1 scores across different classification tasks. Bi-GCN achieved notable improvements, evidenced by an accuracy of 96.1% on the Weibo dataset and similarly high scores on Twitter datasets. These results underscore the model's ability to effectively learn high-level rumor representations.

Implications and Future Directions

Bi-GCN's ability to efficiently model the bi-directional nature of rumors presents significant implications for real-time misinformation management. By incorporating both top-down and bottom-up propagation modeling, Bi-GCN sets a precedent for future research aiming to deal with complex social media data structures. The model's robust early detection capability is particularly essential for timely intervention in misinformation spread.

For future work, expanding this model to incorporate temporal dynamics more explicitly, or extending it to multi-modal data integrating images and text, might provide a more holistic approach to tackling misinformation at scale.

Conclusion

The study illustrates the potential for graph-based neural networks like Bi-GCN to address challenging information dispersal patterns in social media. With promising empirical results and a versatile design, Bi-GCN provides a substantial contribution to the domain of automated rumor detection, paving the way for advancements in social media analytics and broader AI applications.