Supplementary Datasets for "MB-HGTBot: Modeling Multi-Type Social Behaviors in Heterogeneous Graphs for Robust Twitter Bot Detection"

This repository contains the datasets used in the research paper "MB-HGTBot: Modeling Multi-Type Social Behaviors in Heterogeneous Graphs for Robust Twitter Bot Detection". The data is collected and curated to facilitate the study of heterogeneity in user social behavioral patterns and to evaluate the robustness of bot detection models. The collection consists of two main categories: 1. Social Behavior Sub-networks (derived from Twibot-22): These datasets represent heterogeneous information networks constructed based on distinct social interaction types. They allow for the analysis of specific behavioral patterns: Follow: Graph structure based on following behaviors. Reply: Graph structure based on replying behaviors. Like: Graph structure based on liking behaviors. Mention: Graph structure based on mentioning behaviors. Retweet: Graph structure based on retweeting behaviors. Quote: Graph structure based on quote-tweeting behaviors. 2. Topic-Specific Datasets: These real-world datasets are collected to evaluate model generalization and robustness in specific public opinion scenarios: Abortion: Tweets and user data related to abortion discussions. LGBTQ: Tweets and user data related to LGBTQ+ rights discussions. Trump_attacked: Tweets and user data related to discussions surrounding attacks on Donald Trump. Usage: These datasets support the construction of Heterogeneous Graph Transformer models (MB-HGTBot) and are intended for research purposes in social bot detection, graph neural networks, and social network analysis. Related Code: The source code for the model and data preprocessing is available on GitHub: https://github.com/Bemcliu/MB_HGTBot