Learning from evolving data streams: online triage of bug reports

Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, pages 613–622, Avignon, France, April 23 - 27 2012. c⃝2012 Association for Computational Linguistics Learning from evolving data streams: online triage of bug reports Grzegorz Chrupala Spoken Language Systems Saarland University gchrupala@lsv.uni-saarland.de Abstract Open issue trackers are a type of social me- dia that has received relatively little atten- tion from the text-mining community. We investigate the problems inherent in learn- ing to triage bug reports from time-varying data. We demonstrate that concept drift is an important consideration. We show the effectiveness of online learning algorithms by evaluating them on several bug report datasets collected from open issue trackers associated with large open-source projects. We make this collection of data publicly available. 1 Introduction There has been relatively little research to date on applying machine learning and Natural Lan- guage Processing techniques to automate soft- ware project workflows. In this paper we address the problem of bug report triage. 1.1 Issue tracking Large software projects typically track defect re- ports, feature requests and other issue reports us- ing an issue tracker system. Open source projects tend to use trackers which are open to both devel- opers and users. If the product has many users its tracker can receive an overwhelming number of issue reports: Mozilla was receiving almost 300 reports per day in 2006 (Anvik et al. 2006). Some- one has to monitor those reports and triage them, that is decide which component they affect and which developer or team of developers should be responsible for analyzing them and fixing the re- ported defects. An automated agent assisting the staff responsible for such triage has the potential to substantially reduce the time and cost of this task. 1.2 Issue trackers as social media In a large software project with a loose, not strictly hierarchical organization, standards and practices are not exclusively imposed top-down but also tend to spontaneously arise in a bottom- up fashion, arrived at through interaction of in- dividual developers, testers and users. The indi- viduals involved may negotiate practices explic- itly, but may also imitate and influence each other via implicitly acquired reputation and status. This process has a strong emergent component: an in- formal taxonomy may arise and evolve in an is- sue tracker via the use of free-form tags or labels. Developers, testers and users can attach tags to their issue reports in order to informally classify them. The issue tracking software may give users feedback by informing them which tags were fre- quently used in the past, or suggest tags based on the content of the report or other information. Through this collaborative, feedback driven pro- cess involving both human and machine partici- pants, an evolving consensus on the label inven- tory and semantics typically arises, without much top-down control (Halpin et al. 2007). This kind of emergent taxonomy is known as a folksonomy or collaborative tagging and is very common in the context of social web appli- cations. …

Content truncated to 500 words