Automatic Classification of Bug Reports for Mobile Devices: An Industrial Case Study

Abstract

When a failure is found during software testing activities, a bug report (BR) is written and stored in product management tools. In order to prioritize the errors to fix, a BR triage process is performed to identify the most critical errors. This is specifically relevant in the context of mobile applications due to the fast development cycle, which results on a high number of BRs to evaluate daily. In this paper, Machine Learning (ML) and Natural Language Processing (NLP) techniques are investigated to automatically classify the criticality of BRs in the context of a real mobile environment, and a prototype was developed. Results on a corpus of 9,785 BRs were very satisfactory, reaching up to 0.79 of AUC and meeting the performance level required by the considered application.