Optimized Bug Report Triaging Using N-Gram Features and Non-Reproducible Classification

Testing the product before final release is an inevitable part of any engineering process. Software testing is an activity which is carried out with an intent of finding errors. Testing is carried out as an umbrella activity is most modern software development life cycle models. As time-to-market is a critical constraint for most software, the beta version of the software is released with intent of beta testing. It is the testing which is done by the end user as he/she uses the software. Software bugs are corresponding to all the instants when the software does not meet the specifications or desired functionality. In case of software crashes, there is a provision in most of the software that a bug report is automatically generated and sent to a bug reporting or issue tracking system. Also, in case of other failures, the user can write a description of his/her own about the problem faced. All such bug reports must be handled appropriately to remove the corresponding bug. For software which are publicly viable, and freely distributable across the globe, thousands of bug reports are sent to the issue tracking system, probably corresponding to the same bug. A person in-charge of analyzing the bug reports for checking of duplicity and assigning to appropriate developer is called triager. In this paper, the problem of checking of duplicate bug reports is analyzed based on non-repudiation classification and textual features. The non-repudiation classification is important and critical as the reports for which the corresponding bug cannot be reproduced at client side are of no use. The contribution of this work is the classification of bug reports as being reproducible or non-reproducible using a binary classifier. Support Vector Machine is used for this classification. The traditional textual classifier is augmented with N Gram Features to improve the accuracy of classification. N-gram features relates to the character level representation instead of word level representation. For example, "StringIndexOutOfBoundsException" and "StringIndexOutOfBounds" Exception are two sets which possibly relates to the same event. Proposed model is evaluated over Mozilla-Firefox bug repository. Experimental results indicate a 7% improvement in the kappa measure as compared to the independent approaches proposed by Phuc Nhan Minh and Mona Erfani Joorabchi et.al..