A Bidirectional LSTM Log Detection Model Based on Attention Mechanism

Addressing the issue of excessive manpower and time consumption when annotating vast amounts of log data in the process of computing trust values based on log modeling, as well as the problem of reduced accuracy of computed trust values due to lack of attention to log contextual features, this paper proposes a dual-layer Recurrent Neural Network (LSTM) based on the attention mechanism. This model is trained to identify the feature activities of each user on the network and assess whether the user behavior is normal or anomalous. This serves as the foundation for calculating user trust values based on log modeling, ensuring the accuracy of trust value computation. When evaluated using the CERT dataset, the model in this scheme outperforms the control group model, achieving excellent performance metrics on the test dataset.