T-REC: Towards Accurate Bug Triage for Technical Groups

Abstract

With ever-larger software development systems involving more people with different skills, it is necessary to think about the process of bug assignment to groups of developers and not just to a developer alone. This work aims to leverage Bug Triage Process by suggesting a list of specialized groups of developers, or Technical Groups (TG's), to be attributed to a new bug report, based on other bugs that are similar and have been resolved by these TG's in the past. In the dataset used in our experiments, the mean time to correctly assign bug reports to their proper TG is 14 days, and just by then, the bug fixing process starts. This is a critical problem for software development and management since issues tend to accumulate a high-resolution time, which compromises developer performance and deliveries. In order to enhance the Bug Triage Process, we propose T-REC, an auxiliary SW Project Management system that accurately and efficiently analyzes similar issues to provide personalized TG recommendation. T-REC is a method that ensemble Machine Learning (ML) and Information Retrieval (IR) algorithms to recommend a list of TG's to handle an issue. Our experiments show that T-REC recommendation reaches an overall Acc@1 of 50.9%, Acc@2 of 63.2%, Acc@5 of 76.1%, Acc@10 of 83.6%, and Acc@20 of 89.7%. To the best of our knowledge, our work is the first to associate multiple machine learning strategies (classifiers, attributes, and training history) on the prediction of specialized groups of developers. We validate our approach on a real-world dataset from a large company that comprises 9.5M mobile-related bug reports from January 2001 to January 2019.