| dc.description.abstract | Software change requests, such as bug fixes and new features, are an integral part of software evolution and maintenance. It is not uncommon in open source projects to receive numerous change requests daily, which need to be triaged. Therein, automatically assigning the most appropriate developer(s) to resolve an incoming change request is an important task. The thesis proposes two approaches to address this task. The first approach, namely iA, employs a combination of an information retrieval technique and processing of the source code authorship information. The relevant source code files to the textual description of a change request are first located. The authors listed in the header comments in these files are then analyzed to arrive at a ranked list of the most suitable developers. The approach fundamentally differs from its previously reported counterparts, as it does not require software repository mining.
The second approach, namely, iMacPro, amalgamates the textual similarity between the given change request and source code, change proneness information, authors, and maintainers of a software system. Latent Semantic Indexing (LSI) and a lightweight analysis of source code, and its commits from the software repository, are used. The basic premise of iMacPro is that the authors and maintainers of the relevant source code, which is change prone, to a given change request are most likely to best assist with its resolution. iMacPro unifies these sources in a unique way to perform its task, which was not investigated in the literature previously.
An empirical study to evaluate the effectiveness of the approaches on open source systems, ArgoUML, JabRef, jEdit, and MuCommander, is reported. The iA approach is found to provide recommendation accuracies that are equivalent or better than the two compared approaches. Results also show that iMacPro could provide recall gains from 30% to 180% over its subjected competitor with statistical significance. | |
