Embedding and Predicting Software Security Entity Relationships: A Knowledge Graph Based Approach

Software security knowledge involves heterogeneous security concepts (e.g., software weaknesses and attack patterns) and security instances (e.g., the vulnerabilities of a particular software product), which can be regarded as software security entities. Among software security entities, there are many within-type relationships as well as many across-type relationships. Predicting software security entity relationships helps to enrich software security knowledge (e.g., finding missing relationships among existing entities). Unfortunately, software security entities are currently documented in separate databases, such as Common Vulnerabilities and Exposures (CVE), Common Weakness Enumeration (CWE) and Common Attack Pattern Enumeration and Classification (CAPEC). This hyper-document representation cannot support effective reasoning of software entity relationships. In this paper, we propose to consolidate heterogeneous software security concepts and instances from separate databases into a coherent knowledge graph. We develop a knowledge graph embedding method which embeds the symbolic relational and descriptive information of software security entities into a continuous vector space. The resulting entity and relationship embeddings are predictive for software security entity relationships. Based on the Open World Assumption, we conduct extensive experiments to evaluate the effectiveness of our knowledge graph based approach for predicting various within-type and across-type relationships of software security entities.