Automatic bug labeling using semantic information from LSI

Abstract

Most open source projects provide a defect tracking system, where users, developers, testers can directly report the problems. The fields provided in the bug report help triager and debugger to understand the problem better. They also help in other tasks like accurate assessment of priority and severity of bugs, identification of appropriate developer to resolve bugs etc. Label field in the bug report is one such field. It has been observed that in many bug repositories, the label field is either not present or is incorrectly assigned. There is a need for automatic bug labeling so that bug reports could be made more informative. This paper presents an automated technique for bug labeling using TF-IDF and LSI. Experimental study shows that there is improvement in results with the addition of semantically similar words obtained from LSI in conjunction with the terms extracted using TF-IDF. Using LSI along with TF-IDF, we achieved 61.5% accuracy for the polish bug reports and 62.8% accuracy for security bug reports as compared to 53.8% accuracy for polish and 61% for security bug reports from using TF-IDF alone.