Humboldt-Universität zu Berlin - Mathematisch-Naturwissenschaftliche Fakultät - Institut für Informatik

Die Verteidigung findet auch digital per Zoom statt. Eine Zoom-Einladung finden Sie hier. (nur mit Informatik-Account)

Abstract: 

Issue management plays an important role in software engineering. Properly assigning issues to developers and correctly identifying the files that need to be changed to resolve an issue can reduce resolution times and ensure efficient use of resources. This thesis focuses on issue management using machine learning techniques. For this purpose, CNN and Bi-LSTM based two-machine learning models were built for issue assignment. The models aimed to capture similarities between issues and assign new issues to developers who have solved similar issues in the past. Proposed models were evaluated on three different datasets using five-fold time series split and sliding window techniques. In addition, a model was developed for issue localization and tested on a dataset. This model aimed to understand the relationships between the issue description and the data extracted from code files. The proposed models did not provide any improvement to the current state of the art techniques. On the other hand, unlike many other studies, the use of issue labels as input and the consideration of the temporal order of issues in issue assignment models are the important aspects of study.