Optimizing Software Bug Triage Using Deep Learning and Word Embeddings toImprove D evelopers' Productivity

Bug triage is a crucial issue in software maintenance that can impact software costs and requires asuitable tool for assigning bugs to appropriate developers. While many approaches have been developedto automate bug triage, most have only evaluated their methods based on accuracy, precision, recall,and static metrics. They have not considered time-dependent evolutionary metrics, such as workload onexpert developers, task distribution between developers, and delays in bug fixing. Our researchintroduced three deep learning methods that use pre-trained word embedding models: BiLSTM-GloVe,BiLSTM-Word۲Vec, and BERT. We aimed to investigate whether the learning method and wordrepresentation techniques impact bug assignment in static and evolutionary metrics. We first select andpreprocess the Summary and Description of bug reports. Then, we use word embedding techniques torepresent the input data for the neural network. Next, we employ a neural network to calculate theprobability of each developer for bug reports. Finally, the model selects the most suitable developer forbug assignment. Using the Wayback Machine, our methods were evaluated on three well-known opensourceprojects, EclipseJDT, LibreOffice, and Mozilla. The results demonstrate that our approach canenhance evolutionary metrics and optimize developers’ utilization in the bug triage process.

کلیدواژه ها:

Bug triage ، Deep learning ، Word embedding ، Machine learning ، Software maintenance