Heterogeneity-Aware Twitter Bot Detection with Relational Graph Transformers

Abstract

Twitter bot detection has become an important and challenging task to combat misinformation and protect the integrity of the online discourse. State-of-the-art approaches generally leverage the topological structure of the Twittersphere, while they neglect the heterogeneity of relations and influence among users. In this paper, we propose a novel bot detection framework to alleviate this problem, which leverages the topological structure of user-formed heterogeneous graphs and models varying influence intensity between users. Specifically, we construct a heterogeneous information network with users as nodes and diversified relations as edges. We then propose relational graph transformers to model heterogeneous influence between users and learn node representations. Finally, we use semantic attention networks to aggregate messages across users and relations and conduct heterogeneity-aware Twitter bot detection. Extensive experiments demonstrate that our proposal outperforms state-of-the-art methods on a comprehensive Twitter bot detection benchmark. Additional studies also bear out the effectiveness of our proposed relational graph transformers, semantic attention networks and the graph-based approach in general.

Botshape: A Novel Social Bots Detection Approach via Behavioral Patterns

“…Shangbin et al [19] applied GCN algorithm on the user following relationship graph, then represented raw node features including user profile, categorical and numerical data of account activity. Shangbin et al [20] constructs two kinds of heterogeneity structures, including relation and influence, leveraging the topology to identify the difference between genuine users and social bots.…”

Dual-Channel Relation Mixup Strategy for Social Bot Detection Based on Graph Neural Network

“…By constructing heterogeneous graphs and extracting original node features using pre-trained language models to obtain initial node embeddings, followed by aggregation of R-GCN models, BotRGCN [7] demonstrated superior performance compared to conventional detection models on the TwiBot-20 [23], which is a recently released benchmark dataset for social bot detection. Another method proposed in [6] applied heterogeneous graphs and introduced a relational graph converter influenced by natural language processing to model user fusion and learn node representations for improved social bots detection, outperforming BotRGCN. As shown in Fig.…”

“…The latest progress in graph neural networks (GNNs) [5] has facilitated a more comprehensive understanding of the implicit relation between bot users and legitimate users, thereby increasing the complexity of detection. GNN-based methods [6,7] represent the detection process as a node classification problem.…”

Simplistic Collection and Labeling Practices Limit the Utility of Benchmark Datasets for Twitter Bot Detection

“…Leveraging text data, deep learning approaches apply large pre-trained language models [38] or models trained by the researchers themselves [28,39,43,46,48]. Other popular approaches use network features to train graph neural networks [1,20,23] or try to detect botnets [70] using their community structures. Researchers have even sought insight from other disciplines by using behavioral [30,34] or biology-inspired approaches [13][14][15]58].…”