Towards Effective Bug Triage with Software Data Reduction Techniques
Software companies spend over 45 percent of cost in dealing with software bugs. An inevitable step of fixing bugs is bug triage, which aims to correctly assign a developer to a new bug. To decrease the time cost in manual work, text classification techniques are applied to conduct automatic bug triage. In this paper, we address the problem of data reduction for bug triage, i.e., how to reduce the scale and improve the quality of bug data. We combine instance selection with feature selection to simultaneously reduce data scale on the bug dimension and the word dimension. To determine the order of applying instance selection and feature selection, we extract attributes from historical bug data sets and build a predictive model for a new bug data set. We empirically investigate the performance of data reduction on totally 600,000 bug reports of two large open source projects, namely Eclipse and Mozilla. The results show that our data reduction can effectively reduce the data scale and improve the accuracy of bug triage. Our work provides an approach to leveraging techniques on data processing to form reduced and high-quality bug data in software development and maintenance.
A time-consuming step of handling software bugs is bug triage, which aims to assign a correct developer to fix a new bug. In traditional software development, new bugs are manually triaged by an expert developer, i.e., a human triage. Due to the large number of daily bugs and the lack of expertise of all the bugs, manual bug triage is expensive in time cost and low in accuracy. In manual bug triage in Eclipse, percent of bugs are assigned by mistake while the time cost between opening one bug and its first triaging is 19.3 days on average. To avoid the expensive cost of manual bug triage, existing work has proposed an automatic bug triage approach, which applies text classification techniques to predict developers for bug reports. In this approach, a bug report is mapped to a document and a related developer is mapped to the label of the document. Then, bug triage is converted into a problem of text classification and is automatically solved with mature text classification techniques, e.g., Naive Bayes. Based on the results of text classification, a human triager assigns new bugs by incorporating his/her expertise. To improve the accuracy of text classification techniques for bug triage, some further techniques are investigated, e.g., a tossing graph approach and a collaborative filtering approach. However, large-scale and low-quality bug data in bug repositories block the techniques of automatic bug triage. .Since software bug data are a kind of free-form text data (generated by developers), it is necessary to generate well-processed bug data to facilitate the application
In this paper, we address the problem of data reduction for bug triage, i.e., how to reduce the bug data to save the labor cost of developers and improve the quality to facilitate the process of bug triage. Data reduction for bug triage aims to build a small-scale and high-quality set of bug data by removing bug reports and words, which are redundant or non-informative. In our work, we combine existing techniques of instance selection and feature selection to simultaneously reduce the bug dimension and the word dimension. The reduced bug data contain fewer bug reports and fewer words than the original bug data and provide similar information over the original bug data. We evaluate the reduced bug data according to two criteria: the scale of a data set and the accuracy of bug triage. To avoid the bias of a single algorithm, we empirically examine the results of four instance selection algorithms and four feature selection algorithm.
- We present the problem of data reduction for bug triage. This problem aims to augment the data set of bug triage in two aspects, namely
- To simultaneously reduce the scales of the bug dimension and the word dimension.
- to improve the accuracy of bug triage.
- We propose a combination approach to addressing the problem of data reduction. This can be viewed as an application of instance selection and feature selection in bug repositories.
- We build a binary classifier to predict the order of applying instance selection and feature selection. To our knowledge, the order of applying instance selection and feature selection has not been investigated in related domains.
In this part, we present the data preparation for applying the bug data reduction. We evaluate the bug data reduction on bug repositories of two large open source projects, namely Eclipse and Mozilla. Eclipse is a multi-language software development environment, including an Integrated Development Environment (IDE) and an extensible plug-in system; Mozilla is an Internet application suite, including some classic products, such as the Firefox browser and the Thunderbird email client. Up to December 31, 2011, 366,443 bug reports over 10 years have been recorded to Eclipse while 643,615 bug reports over 12 years have been recorded to Mozilla. In our work, we collect continuous 300,000 bug reports for each project of Eclipse and Mozilla, i.e., bugs 1-300000 in Eclipse and bugs 300001600000 in Mozilla. Actually, 298,785 bug reports in Eclipse and 281,180 bug reports in Mozilla are collected since some of bug reports are removed from bug repositories (e.g., bug 5315 in Eclipse) or not allowed anonymous access (e.g., bug 40020 in Mozilla). For each bug report, we download web pages from bug repositories and extract the details of bug reports for experiments.
Since bug triage aims to predict the developers who can fix the bugs, we follow the existing work to remove unfixed bug reports, e.g., the new bug reports or will-not-fix bug reports. Thus, we only choose bug reports, which are fixed and duplicate (based on the items status of bug reports). Moreover, in bug repositories, several developers have only fixed very few bugs. Such inactive developers may not provide sufficient information for predicting correct developers. In our work, we remove the developers, who have fixed less than 10 bugs.
Word dimension: We use feature selection to remove noisy duplicate words in a data set
Bug dimension: Instance selection can remove uninformative bug reports, meanwhile, we can observe that the accuracy may be decreased by removing bug reports.
Word dimension: By removing uninformative words, feature selection improves the accuracy of bug triage. This can recover the accuracy loss by instance selection.
Instance selection and feature selection are widely used techniques in data processing. For a given data set in a certain application, instance selection is to obtain a subset of relevant instances (i.e., bug reports in bug data) while feature selection aims to obtain a subset of relevant features (i.e., words in bug data). In our work, we employ the combination of instance selection and feature selection. To
- In our work, to save the labor cost of developers, the data reduction for bug triage has two goals.
1) reducing the data scale.
2) improving the accuracy of bug triage.
- In contrast to modelling the textual content of bug reports in existing work, we aim to augment the data set to build a preprocessing approach, which can be applied before an existing bug triage approach. We explain the two goals of data reduction as follows.
Bug triage is an expensive step of software maintenance in both labor cost and time cost. In this paper, we combine feature selection with instance selection to reduce the scale of bug data sets as well as improve the data quality. To determine the order of applying instance selection and feature selection for a new bug data set, we extract attributes of each bug data set and train a predictive model based on historical data sets. We empirically investigate the data reduction for bug triage in bug repositories of two large open source projects, namely Eclipse and Mozilla. Our work provides an approach to leveraging techniques on data processing to form reduced and high-quality bug data in software development and maintenance. In future work, we plan on improving the results of data reduction in bug triage to explore how to prepare a high quality bug data set and tackle a domain specific software task. For predicting reduction orders, we plan to pay efforts to find out the potential relationship between the attributes of bug data sets and the reduction orders.