Automated Duplicate Defect Detection for Bug Tracking Systems

Abstract

A bug tracking system (BTS) keeps track of the status of a software system in real time. The bug report it generates is sent to the software developer or centre for maintenance whenever it identifies an abnormal scenario. The freshly reported defect, on the other hand, could be a repeat in the master report repository of an old issue with a remedy already present. This situation results in an onslaught of replicate reports of bugs, making the software development life cycle difficult to manage. As a result, in the software industry, it is now an essential task to find repeat reports of bugs early. This research proposes a two-tier method based on topicbased clustering done by LDA approach, multimodal representation of text using W2V, FT, GloVe and a measure of unified text similarity utilizing similarities of the Cosine and Euclidean nature to solve this challenge. The Eclipse dataset, which contains over 80,000 bug reports and includes both master and duplicate reports, is used to validate the suggested method. This investigation focuses primarily on the report descriptions in order to identify duplication. For Top-N proposals, the recommended two-tier technique has achieved a 75% recall rate, which is higher than the traditional one-on-one classification model.