Improving Cybersecurity Named Entity Recognition with Large Language Models
Zhe Qiao, Chen Zhang, Gang Du · 2023
A lot of attention has been paid to cybersecurity threat intelligence analysis based on security knowledge graphs because they can evaluate multi-source threat intelligence data at a fine-grained level. The primary effort in building knowledge graphs is named entity recognition (NER). However, the retrieved features from classic NER algorithms are insufficient to identify novel security entities in the cybersecurity field. Currently, the natural language processing(NLP) field has made unprecedented progress, driven by large language models(LLMs). In this paper, we present a BERT- BiLSTM-CRF cybersecurity NER method based on BiLSTM-CRF that combines the pre-trained large language model BERT. The preprocessed annotated corpus is fed into the Bi-LSTM and CRF models, which employ the word vectors to extract contextual features and annotate 9 different categories of named entities. The proposed cyberecurity NER method is effective in identifying cybersecurity entities on large-scale cybersecurity datasets, and the relevant evaluation indexes are superior to other algorithms, according to comparative experiments carried out within the cyberspace security corpus.