Ontology Learning and Its Application in Software-Intensive Projects

Software artifacts, such as requirements, design, source code, documentation, and safety-related artifacts are typically expressed using domain-specific terminology. Automated tools which attempt to analyze software artifacts in order to perform tasks such as trace retrieval and maintenance, domain analysis, program comprehension, or to service natural language queries, need to understand the vocabulary and concepts of the domain in order to achieve acceptable levels of accuracy. Domain concepts can be captured and stored as an ontology. Unfortunately, constructing ontologies is extremely time-consuming and has proven hard to automate. This dissertation proposes a novel approach for semi-automated ontology building that leverages user-defined trace links to identify candidate domain facts. It uses a variety of web-mining, Natural Language Processing, and machine learning techniques to filter and rank the candidate facts, and to assist the user in building a domain-specific ontology. The benefits of the constructed ontology are described and evaluated within the context of automated trace link creation.