Mining Relations from Git Repositories- Applying Relation Extraction Technology to Git Commit Messages

Abstract

Text data can contain valuable information that is unavailable at a larger scale due to the unstructured nature of free text. Git repositories and Git commit messages within them are one such category of unstructured text data. Relation Extraction (RE) has enjoyed success as a solution to similar problems for a more generic case but also for more specialized domains such as life sciences. RE does however, remain largely untested for text data from Git repositories. This thesis contributes to RE and Software Engineering research by testing RE solutions developed for the generic problem on the domain speci c problem of Git commit messages. An experiment is conducted where a custommade relation extractor is tested on hand annotated Git commit messages drawn from popular public projects on GitHub. The results show that common RE solutions and their models cannot be directly applied to data from Git commit messages due to a very domain spec c language in which these messages are expressed. This prompts for future e orts into developing domain speci c tools and models.