Optimizing Automated Triaging using Contextual Information and Feature Extraction

As software systems are getting larger and complex day by day, software bugs are inevitable. A software crash is one of the most severe manifestations of a defect (bug) in software, and is typically assigned a high priority to be fixed. To facilitate debugging process, many crash reporting systems such as Windows Error Reporting, Apple crash report, and Mozilla crash report have been deployed to automatically collect crash reports from users at the time of crash. Bug reports can also accompany other malfunctions of the software, mostly for the beta or unstable versions of the software. Most often, these bug reports are augmented with user contributed experiences as to what actually faced by him/her. Bugs occur for a variety of reasons, ranging from ill-defined specifications, to carelessness, to a programmers misunderstanding of the problem, technical issues, non-functional qualities, corner cases, etc. Addressing these bugs frequently accounts for the majority of effort spent in the maintenance phase of a software project's life-cycle. This is why, researchers have been trying to enhance the bug-tracking systems to facilitate the bug-fixing process. The person who is in charge of processing the newly reported bugs, checking for duplicates and passing them to appropriate developers to get fixed is called a triager and this process is called triaging. In this paper, an model of automated triaging process is proposed based on contextual information and bug report textual similarity features. In proposed model, the preprocessed bug reports are analyzed for contextual information, to relate them to the non-functional requirements of the software. This extends the feature extraction a step further. The weighted sum of similarity score of all the features, including textual and contextual, is evaluated and used to classify bug reports. Positive and negative sets of existing classification are used to train SVM model. Simulation is done using R simulation package and the results shows a considerable improvement against classification without contextual information.