Improving Quality of Programming and Software through Knowledge Graph Construction and Application

Abstract

Knowledge Graph has been widely used in different domains, including product recommendation, and searching engine. Such application have brought huge convenience to people lives. People can conveniently find relevant information through the google search, including the relevant people or incidents. The powerful product recommendation can also benefit the shopping platforms like Alibaba and eBay, creating great values each year. However, although knowledge graph has played an important role in such achievements, there are few works focusing on using such technique in software domain, except us. In this dissertation, we discuss our findings of knowledge burying issues among different types of programming online tutorials, including API usage directives and task programming materials. Because of poor document design, online programming tutorials that contain abundant knowledge are often ignored by developers. We discover that extract key ontologies from the tutorials and construct knowledge graph can support question answering and knowledge recommendation, which is able to relief the knowledge burying problems. Consequently, we proposed methods constructing API usage directive knowledge graph and task-oriented programming knowledge graph support knowledge retrieval. Our experiments prove the high accuracy and efficiency of our construction methods and the designed user studies prove the usefulness of our knowledge graphs. Besides, we also discover with sheer amount of software vulnerabilities, information discrepancy issues among different security platforms become severe, leading to obstacles for integral knowledge application. To fully understand the discrepancy issues and relief the problem for better knowledge usage, we conduct an empirical study about vulnerability key aspect discrepancy issues among four carefully selected representative vulnerability platforms. We also propose several methods for traceability recovery and silent fix detection for the data integration. Our experiments prove the effectiveness of our methods, which contribute to the vulnerability knowledge conformity and lay the foundations for vulnerability knowledge graph construction for better knowledge usage.