Issue Links Retrieval for New Issues in Issue Tracking Systems

Issue Tracking Systems (ITSs) are used to manage software issues, such as ‘feature requests’ and ‘bug’ reports. Issues are often interconnected by labelled links to show the relationships between issues. Recent studies in duplicate bug report retrieval (DBRR) aim to retrieve relevant duplicate bug reports o for a query bug report s to form duplicate pairs (s, o). However, the DBRR task has its limitations: (1) only focuses on the issues with the ‘bug’ type and overlooks other issue types, e.g., ‘task’ and ‘improvement’; (2) only issues with respect to the ‘duplicate’ link type are retrieved, whereas a query s may have other link types, e.g., ‘relates’ and ‘contains’, with candidate issues o. This paper goes beyond the DBRR task and proposes a new task of issue link retrieval (ILR) for any issue types and link types. For a given query s, the ILR task will locate all relevant tail issues o with respect to a given link type r to build issue links (s, r, o). This paper presents novel methods using pre-trained language models (e.g., GPT-2) to embed issue links for computing probability scores to retrieve issue links. Our methods are evaluated on four datasets, demonstrating high effectiveness in issue link retrieval. The code and datasets of this paper can be obtained from GitHub ( https://github.com/MiaoHu-Pro/ILR ).