A Text Mining Based Method For Cyber Vulnerability Categorization

With the rapid development of computing and information technology,computer system has become the most common asset in all companies andvarious industries. As a result, the information security issue has evolvedinto a key research field in last decades. Many studies have proved thatcombination of exploits is typical means to compromise a network system.How to manage the rapidly increased number of cyber vulnerabilities hasbecome a challenging job.This paper presents an intelligent method for analyzing and classifyingvulnerabilities based on binary tree supported vector machine after a deepresearch on the existing vulnerability managing solutions. The main work ofthis thesis includes:1) Presenting an intelligent method for analyzing and classifyingvulnerabilities based on binary tree supported vector machine andimproving the algorithm of constructing decision tree for categorization.2) Developing a prototype system which consists of threeinterdependent parts: a security management oriented vulnerability scanner,a model of collecting and preprocessing vulnerability related text resources,and a server used for machine learning and predicting vulnerability cases.3)2742vulnerabilities published recently by CERT are adopted for aseries of empirical test which demonstrate great efficiency of vulnerabilitycategorization. The average accuracy of test was proved to be84.4%.This research can greatly reduce the human effort of vulnerabilitycategorization and management. The results generated by this study can beapplied to detecting multistage attack, correlating intrusion alerts, andgenerating attack graph which indicate a considerable significance of research and application.