Research on quality assessment methods for cybersecurity knowledge graphs

In light of the continuous evolution of cyber threats and the escalating frequency of cyberattacks, cybersecurity knowledge graphs have emerged as vital tools for comprehending and countering cyber threats. Nonetheless, the quality of these knowledge graphs is of paramount significance for their effectiveness in cybersecurity applications. This research proposes a Parameter-Efficient Knowledge Graphs Quality Assessment model (PEKGQA), aimed at evaluating the quality of cybersecurity knowledge graphs. We partition the knowledge graph into different communities and select reserved entities based on degree centrality within each community, and we only learn the embeddings of these reserved entities, significantly reducing the model's parameter size. Moreover, we leverage linkage relationships information and K-nearest neighbor information to encode all entities. Then we employ graph attention networks for the joint iterative update of entity and relation embeddings. This approach not only reduces the parameter count but also achieves excellent quality assessment results. Experimental outcomes demonstrate that the PEKGQA model-based approach significantly outperforms traditional methods in evaluation metrics such as F1 value and accuracy, providing an efficient and accurate solution for quality assessment in the field of cybersecurity knowledge graphs.