Research on Parameter-Efficient Knowledge Graph Completion Methods and Their Performance in the Cybersecurity Field

As an advanced knowledge management and reasoning tool, knowledge graph technology can significantly enhance the efficiency of managing and analyzing cyberthreat intelligence, providing strong technical support for cybersecurity threat identification and situational awareness. This paper proposes a multi-level graph clustering-based model for completing cyberthreat intelligence knowledge graphs (GCCKG). The model utilizes a multi-level graph clustering approach to divide the knowledge graph into communities and selects a portion of high-degree entities within each community as the reserved entities. Additionally, we propose a path-enhanced cosine similarity measurement method to measure the similarity between entities, and a graph attention network is employed to iteratively update the embeddings of entities and relations, capturing key information in the graph-structured data. Experimental results demonstrate that the proposed GCCKG model significantly improves evaluation values such as MRR and Hits@K in knowledge graph completion tasks on several general domain knowledge graphs, including FB15K, WN18, FB15K-237, as well as the cyberthreat intelligence knowledge graph CS13K, while also significantly reducing the model’s parameter size. This provides a novel solution for knowledge graph completion in the field of cyberthreat intelligence.