Abstract

To address the limitations of traditional methods in information system vulnerability management, including low efficiency, high false positive rates, and weak generalization ability, this paper proposes a knowledge graph-driven framework for vulnerability identification and intelligent management. At the core of this framework is ProConTPLinker, an end-to-end entity-relationship extraction model that integrates prompt learning and contrastive learning, specifically designed to resolve the issues of blurred entity boundaries and entity overlapping in vulnerability texts, especially in few-shot scenarios. Experimental results show that ProConTPLinker achieves an F1-score of 0.751, outperforming mainstream algorithms such as CasRel and TPLinker. Building this model, this paper constructs a vulnerability knowledge graph and implements an intelligent management system, verifying the practical application value of the proposed model. This research provides an effective technical solution for information system security construction in government, finance, education, and other industries.

Download

Share

COinS