SeBot: Structural Entropy Guided Multi-View Contrastive Learning for Social Bot Detection

TL;DR

SeBot tackles social bot detection on multi-relational social graphs by integrating structural entropy-guided encoding trees to reveal hierarchical communities, a heterophily-aware encoder to handle adversarial bots, and multi-view self-supervised contrastive learning to unify diverse representations. It introduces three coupled representations from different views, optimizes them with a combined objective $\,\mathcal{L}=\mathcal{L}_{CE}+\lambda_1\mathcal{L}_{NCL}+\lambda_2\mathcal{L}_{SCL}$, and demonstrates superior performance over state-of-the-art baselines on TwiBot-20 and MGTAB. The work also provides thorough ablations, sensitivity analyses, and visualizations to validate the contributions of encoding trees, relational aggregation, and contrastive learning, highlighting robustness to adversarial behavior and improved generalization for real-world deployment.

Abstract

Recent advancements in social bot detection have been driven by the adoption of Graph Neural Networks. The social graph, constructed from social network interactions, contains benign and bot accounts that influence each other. However, previous graph-based detection methods that follow the transductive message-passing paradigm may not fully utilize hidden graph information and are vulnerable to adversarial bot behavior. The indiscriminate message passing between nodes from different categories and communities results in excessively homogeneous node representations, ultimately reducing the effectiveness of social bot detectors. In this paper, we propose SEBot, a novel multi-view graph-based contrastive learning-enabled social bot detector. In particular, we use structural entropy as an uncertainty metric to optimize the entire graph's structure and subgraph-level granularity, revealing the implicitly existing hierarchical community structure. And we design an encoder to enable message passing beyond the homophily assumption, enhancing robustness to adversarial behaviors of social bots. Finally, we employ multi-view contrastive learning to maximize mutual information between different views and enhance the detection performance through multi-task learning. Experimental results demonstrate that our approach significantly improves the performance of social bot detection compared with SOTA methods.

Figures (7)

• Figure 1: Illustration of community structure and inter-class interactions co-exists in social networks. The abstraction of the hierarchical community structure is presented in the form of an encoding tree.
• Figure 2: Overview of our proposed framework SeBot, which mainly consists of three modules: 1) Node-level encoding tree generation and bottom-up message passing; 2) Subgraph-level encoding trees generation and message-passing; 3) Relational information aggregation beyond homophily. Contrastive learning loss and classification loss are later calculated on obtained tree types of representations.
• Figure 3: Sensitive analysis of hyperparameter $\lambda_1$ and $\lambda_2$ on TwiBot-20.
• Figure 4: Account representations visualization on TwiBot-20. Red represents bots, while blue represents humans.
• Figure 5: A case study of local community structure and generated edge attention. The same background color represents belonging to the same sub-community, with the same parent node on the constructed encoding tree.