BiTA: Bidirectional Gated Recurrent Unit-Transformer Aggregator in a Temporal Graph Network Framework for Alert Prediction in Computer Networks

BiTA proposes a bidirectional gated recurrent unit-transformer aggregator for temporal graph neural networks to improve alert prediction in computer networks. The method captures recursive, multi-scale temporal patterns in network attack behaviors through bidirectional sequential dependencies and long-range contextual relations. Evaluation demonstrates significant improvements in key metrics across real-world alert datasets compared to state-of-the-art temporal graph models.