Software Defects Automatically Assigned

Nowadays, with the increasing of size and function requirements in big soft-ware system, there are more and more bugs be discovered and reported to the maintainers. Manually assign so many bug reports to developers are time cost and boring. In order to manage the bugs effectively, many big systems build cor-responding bug tracking systems, which record the information and processing of each bug. This information provides the possibility of automatic processing of bug triage. The purpose of Automatic Bug Triage is to assign a newly reported bug to some fixer by using the history bug reports and data mining technologies.Traditional bug triage methods are text classification based, which treat each bug report as a document, and label the document with those who had fixed the bugs, and then classifiers are trained by these history data. Each time a new bug report is submitted, it will be assigned to a fixer through these classifiers.Comparing with classical text categorization problem, bug triage problem has some other characteristics. For example, the description information in bug report is always simple and short, and the number of developers as labels is large. At the same time, there are also some other auxiliary information, which can be used to improve the bug triage accuracy, such as the platform and component of a bug, the relationship between developers, and so on.Focus on the automatic bug triage problem, this paper makes a survey of its background and related works, at the same time, we summarizes the main characteristics of bug triage problem. Inspired by the high dimensionality, sparse-ness and noise containing characteristics of word feature space, we proposes an PLSA/LDA topic model based bug triage method, which maps the bug reports to low dimensional topic space, and then makes bug triage. Experiment results show that, our method can increase the accuracy of bug triage, and obviously reduces the dimension of feature space.