No abstract

Duplicate Defect Detection Tomi Prifti Discovering and fixing faults is an unavoidable process in Software Engineering. It is always a good practice to document and organize fault reports. This facilitates the effectiveness of development and maintenance process. Bug Tracking Repositories, such as Bugzilla, are designed to provide fault reporting facilities for developers, testers and users of the system. Allowing anyone to contribute finding and reporting faults has an immediate impact on software quality. However, this benefit comes with one side-effect. Users often file reports that describe the same fault. This increases the triaging time spent by the maintainers. At the same time, important information required to fix the fault is likely to be distributed across different reports. The objective of this thesis is twofold. First, we want to understand the dynamics of bug report filing for a large, long duration open source project, Firefox. Second, we present a new approach that can reduce the number of duplicate reports. The novel element in the proposed approach is the ability to concentrate the search for duplicates on specific portions of the bug repository. This improves the performance of Information Retrieval techniques and classification runtime of our algorithm. Our system can be deployed as a search tool to help reporters query the repository or it can be adopted to help maintainers detect duplicate reports. In both cases the performance is satisfactory. When tested as a search tool our system is able to detect up to 53% of duplicate reports. The approach adapted for maintainers has a maximum recall rate of 59%.