A Chinese Cyber Threat Intelligence Named Entity Recognition Method
Jinfu Song · 2025
Chinese cybersecurity named entity recognition often suffers from model performance problems due to data noise and insufficient feature extraction. To address this problem, this paper proposes the BERT-ABIM-CRF model, which uses BERT to extract dynamic word embeddings optimized by adversarial training, combines BiLSTM and IDCNN for joint feature extraction, and fuses global and local features using a multi-head self-attention mechanism. Finally, CRF is used for decoding to ensure the consistency of the results. Experiments show that the model has an F1 score of 0.8305 on the CDTier dataset and 0.9122 on the cybersecurity dataset, and exhibits strong robustness in dealing with noise and long-range dependencies. This greatly improves the recognition accuracy and performance, validating its potential in the field of cyber threat intelligence.