Automated software issue triage in large scale industrial context

Official URL: https://risc01.sabanciuniv.edu/record=b2579291_(Table of contents)

Software issue reports are the documents describing the problems users face when using a software product and software issue triage is the process of validating and assigning these issue reports. In practice, issue triage is carried out manually by experts or developers. In large scale industrial contexts, hundreds of software products exist and hundreds of issue reports are filed every day. It takes a great amount of human effort to triage these reports and failure to solve them on time results in customer dissatisfaction. In this thesis, we automate the issue triage process by using data mining approaches and share our experience gained by deploying the resulting system in a large scale industrial setting. Deployment of such a system presented us not only with an opportunity to observe the practical effects of automation, but also to carry out user studies, both of which have not been done before in this context. Furthermore, we developed and empirically evaluated methods on how to create human-readable, non-technical explanations for the predictions made, and on how to monitor and detect deteriorations in accuracies in an online manner. In our efforts to improve the performance, we analyzed the incorrectly assigned issue reports. We realized that many of them have attachments with them, which are mostly screenshots, and such reports generally have short or insufficient descriptions for the problem. Based on these observations, we further carried out studies on how to ensure that we detect the missing information in the descriptions of issue reports automatically and how we can use the attached screenshots as an additional source of information, in order to improve the performance of the automation.