An Information Completion Method Based on Security Knowledge Graph with Fusing Neighborhood Information

In the research of knowledge graph completion, the issue of missing information in open-source network security knowledge repositories has persisted due to challenges such as difficulty in coordinating heterogeneous information and maintaining historical data. To address the problem of insufficient feature learning from different domains in existing information completion methods, a information completion method based on security knowledge graph with fusing neighborhood information (KGC-N), is proposed. To capture neighborhood information, this method constructs a security knowledge graph associated with the Structured Threat Information Expression (STIX 2.1) open-source network security knowledge repository. During the data preprocessing phase, Global Vectors for Word Representation (GloVe) is used to enhance the model's training effectiveness. Leveraging graph traversal techniques, it captures neighborhood information with various relationships within a multi-hop range through the knowledge graph. The captured neighborhood features are then learned using a Graph Attention Network to enhance the model's predictive performance. Experimental results indicate that KGC-N achieves a Mean Ranking of 196 and a Mean Reciprocal Ranking of 0.6127, outperforming baseline methods.