BotSCL: Heterophily-aware Social Bot Detection with Supervised Contrastive Learning

TL;DR

This paper tackles the problem of social bot detection in networks where bots and humans interact heterophilically, which degrades many graph-based detection methods. It introduces BotSCL, a heterophily-aware framework that uses two graph augmentations to create distinct views and a channel-wise encoder guided by supervised contrastive learning to learn class-consistent representations across views. The approach jointly differentiates heterophilic neighbor information while aggregating homophilic signals, and uses a cross-view contrastive loss to improve generalization. Experiments on TwiBot-20 and TwiBot-22 show BotSCL achieves state-of-the-art accuracy and F1, with ablations confirming the critical roles of supervision, augmentation choices, and multi-relational modeling. The results establish a robust, scalable framework for detecting evolving social bots in real-world networks.

Abstract

Detecting ever-evolving social bots has become increasingly challenging. Advanced bots tend to interact more with humans as a camouflage to evade detection. While graph-based detection methods can exploit various relations in social networks to model node behaviors, the aggregated information from neighbors largely ignore the inherent heterophily, i.e., the connections between different classes of accounts. Message passing mechanism on heterophilic edges can lead to feature mixture between bots and normal users, resulting in more false negatives. In this paper, we present BotSCL, a heterophily-aware contrastive learning framework that can adaptively differentiate neighbor representations of heterophilic relations while assimilating the representations of homophilic neighbors. Specifically, we employ two graph augmentation methods to generate different graph views and design a channel-wise and attention-free encoder to overcome the limitation of neighbor information summing. Supervised contrastive learning is used to guide the encoder to aggregate class-specific information. Extensive experiments on two social bot detection benchmarks demonstrate that BotSCL outperforms baseline approaches including the state-of-the-art bot detection approaches, partially heterophilic GNNs and self-supervised contrast learning methods.

Figures (5)

• Figure 1: Illustration of Heterophily Influence. Message passing on heterophilic edges leads to feature mixing and classification boundary shift in social bot detection.
• Figure 2: The proposed BotSCL framework.
• Figure 3: Heterophily influence on previous graph-based methods.
• Figure 4: Sensitive Analysis of Hyperparameter $\lambda^{\{1\}}$ and $\lambda^{\{2\}}$.
• Figure 5: User Representations Visualization. Red represents bots, while blue represents humans.