Identifying social bots via heterogeneous motifs based on Naïve Bayes model

TL;DR

This work tackles social bot detection by modeling neighborhood heterogeneity through heterogeneous motifs derived from node-labels and a Naïve Bayes framework to quantify motif-level contributions. It introduces 114 heterogeneous 3-node motifs and a maximum capability measure to guide feature selection, demonstrating that a subset of high-capability motifs can approach the performance of using all motifs. The approach yields state-of-the-art results on four large benchmarks using only network topology, and exhibits robustness across multiple classifiers (XGBoost, Random Forest, Gradient Boosting). By providing an interpretable, theoretically grounded pipeline, the method offers a scalable, privacy-friendly alternative to deep learning models while suggesting avenues for integrating higher-order structures in future work.

Abstract

Identifying social bots has become a critical challenge due to their significant influence on social media ecosystems. Despite advancements in detection methods, most topology-based approaches insufficiently account for the heterogeneity of neighborhood preferences and lack a systematic theoretical foundation, relying instead on intuition and experience. Here, we propose a theoretical framework for detecting social bots utilizing heterogeneous motifs based on the Naïve Bayes model. Specifically, we refine homogeneous motifs into heterogeneous ones by incorporating node-label information, effectively capturing the heterogeneity of neighborhood preferences. Additionally, we systematically evaluate the contribution of different node pairs within heterogeneous motifs to the likelihood of a node being identified as a social bot. Furthermore, we mathematically quantify the maximum capability of each heterogeneous motif, enabling the estimation of its potential benefits. Comprehensive evaluations on four large, publicly available benchmarks confirm that our method surpasses state-of-the-art techniques, achieving superior performance across five evaluation metrics. Moreover, our results reveal that selecting motifs with the highest capability achieves detection performance comparable to using all heterogeneous motifs. Overall, our framework offers an effective and theoretically grounded solution for social bot detection, significantly enhancing cybersecurity measures in social networks.

Figures (4)

• Figure 1: The 30 unique 3-node homogeneous motifs in a directed network. The yellow node is the target node to be detected. A) 6 first-order homogeneous motifs. B) 15 closed homogeneous motifs. C) 9 second-order homogeneous motifs.
• Figure 2: The heterogeneous motifs by refining homogeneous motifs with node-label information. The yellow node is the target node to be detected. The green and red nodes are human and bot, respectively.
• Figure 3: An example demonstrating the differing contributions of node pairs $(B, C)$ and $(D, E)$ to the probability of $A$ being identified as a social bot. The node $A$ is the target node under evaluation. For heterogeneous motif Y1, the node pairs $(B, C)$ along with nodes $F$ and $G$ can successfully form its motif Y1. In contrast, the node pairs $(D, E)$ are unable to form motif Y1, highlighting their differing roles in the detection process.
• Figure 4: The maximum capability of a heterogeneous motif. The horizontal axis is the maximum capability of a heterogeneous motif. The vertical axis is the real AUC performance of a heterogeneous motif by the XGBoost classifier. Note that each plot contains 114 data points. In the case of Cresci-15, most heterogeneous motifs have both a maximum capability and an actual AUC of 0.5, resulting in substantial overlap among the points.