Research On Entity Relationship Extraction Of Cyber Security Knowledge

Internet text has become an important carrier of cyber security information on the Internet.These texts not only include the latest security incidents,such as exploits,data leaks,and hacker attacks,but also include a large amount of security knowledge,such as security blog forums,cyber security knowledge platforms,etc.By integrating large-scale,heterogeneous,and unstructured cyber security information,and extracting entity relationships from cyber security information,it is possible to effectively master cyber security knowledge and assist in the resolution of cyber security issues.On publicly accessible cyber security datasets,experimental findings show that the proposed approach outperforms other approaches in extracting cyber security entities more effectively.Due to the complexity and diversity of texts in the field of cyber security,traditional named entity recognition methods are difficult to identify cyber security entities,and the research in this field is still in its infancy.Aiming at the difficulties of entity recognition in the field of cyber security,this paper proposes a new model,Bi LSTM-DIC-ATT-CRF,which combines a knowledge-driven dictionary system with a data-driven deep learning method.On the public cyber security dataset,the experimental results indicate that the approach introduced in this paper outperforms other approaches,and can extract cyber security entities more effectively.Based on the recognition of named entities in cyber security,the relationship extraction of cyber security knowledge is carried out.Aiming at the problem of triple overlap in relation extraction,this paper proposes a joint extraction method of entity relationships based on relation decomposition.The model mainly contains three modules:encoder module,relation extraction module and entity recognition module.The input text is firstly generated by the encoder module to generate a text word vector representation,and then the obtained text word vector is pooled to reduce the dimensionality,so as to obtain the text sentence vector representation.In addition,the attention mechanism is introduced in the process of sentence vector generation to capture the importance of different words to sentence classification.By combining the two to obtain a new text sentence vector,multi-relation classification is performed in the relation extraction module.Finally,the specific relationship vector and the text word vector are combined,and the entity recognition module recognizes the entities under the specific relationship,thereby generating triples.Experimental results show that the proposed model effectively solves the problem of triple overlap and improves the performance of relation extraction.