Automated Data Augmentation For Entity Alignment In Vulnerability Knowledge Graphs

The purpose of knowledge graph fusion is to match the corresponding entities and relationships in knowledge graphs from different builders in each domain to obtain a more complete and richer knowledge graph.However,due to the subjectivity of knowledge graph builders and the duality of knowledge,there are often entities with different representations but the same meaning in different graphs,and the goal of entity alignment task is to discover such entities.Entity alignment is one of the most important tasks in knowledge graph fusion,and therefore has been widely studied.However,most of the current entity alignment methods focus on generic encyclopedic graphs,and the entity alignment methods on domain graphs have not been sufficiently studied.In the field of cybersecurity,vulnerability knowledge graphs are a typical class of domain knowledge graphs,and how to align entities in vulnerability graphs from different sources is a necessary task in the construction of cybersecurity knowledge graphs.However,vulnerability graphs have their unique structural characteristics,and the direct application of current generic methods is often ineffective because the structural information is not rich and diverse enough.To this end,this thesis conducts research on data enhancement methods for the task of entity alignment of vulnerability maps,aiming to provide richer and more diverse structural information for entity alignment through data enhancement,so as to improve the effect of entity alignment.The specific work is as follows.(1)Entity alignment study of vulnerability mapping: The vulnerability knowledge map is constructed based on CNNVD and CNVD,and the effect of entity alignment is analyzed based on the generic method by using the structural information of the map on it.The results show that RDGCN can achieve the best alignment effect among the generic methods,but the accuracy is still not satisfactory,and special mechanisms need to be designed for vulnerability mapping to serve the entity alignment task.(2)Data enhancement method for vulnerability map entity alignment: Based on the in-depth analysis of vulnerability map structure,three data enhancement methods are proposed: relationship redefinition,structural linkage enhancement and attribute ontology selection,which gradually expand the relationship types and the number of relationship triples of vulnerability map and enhance the structural information of vulnerability map.The experimental results prove that after data enhancement,the entity alignment effect of vulnerability mapping is significantly improved.(3)Automated data enhancement method for vulnerability map entity alignment:In order to reduce the manual workload in the process of data enhancement,combining three evaluation indexes of ontology diversity,entity number diversity and attribute ontology relevance,we propose to conduct data enhancement of vulnerability map in an automated way,and automatically filter out the ontologies and entities in the vulnerability map that are suitable to participate in entity alignment work.The experiments show that the automated data augmentation proposed in this paper does not require manual participation and can effectively improve the accuracy of the entity alignment results of the vulnerability map.