Bot-MGAT: A Transfer Learning Model Based on a Multi-View Graph Attention Network to Detect Social Bots

Twitter, as a popular social network, has been targeted by different bot attacks. Detecting bots is challenging task, due to their evolving capacity avoid detection. Extensive research efforts have proposed techniques and approaches solving this problem. Due the scarcity of recently updated labeled data, performance detection systems degrades when exposed new dataset. Therefore, semi-supervised learning (SSL) can improve performance, using both unlabeled examples. In paper, we propose framework based on multi-view graph attention mechanism transfer (TL) approach, predict bots. We called ‘Bot-MGAT’, which stands for network. The used data. profile features reduce overheads feature engineering. executed our experiments recent benchmark dataset that included representative samples with structural information only. applied cross-validation uncertainty in model’s performance. Bot-MGAT was evaluated SSL techniques: single networks (GAT), convolutional (GCN), relational (RGCN). compared related work field results TL outperformed, an accuracy score 97.8%, F1 0.9842, MCC 0.9481.