A feature location approach supported by time-aware weighting of terms associated with developer expertise profiles

Feature location is a frequent software maintenance activity that aims to identify initial source code location pertinent to a software feature. Most of feature location approaches are based, at least in part, on text analysis methods which originate from the natural language context. However, the natural language context and the text data in software repositories have different properties that reveal the need for adaption of the methods to apply in the context of software repositories. One of the differences is the existence of a set of metadata, such as developer information and time stamp, which is associated with the data in the repositories. However, this difference has not been fully considered in previous feature location research studies. This study proposes a feature location approach that analyzes developer expertise profiles, which contain source code entities modified by the associated software developers, to identify the most similar location pertinent to a desired feature. This approach uses a time-aware term-weighting technique to determine the similarity. An experimental evaluation on four open-source projects shows an improvement in the accuracy, performance, and effectiveness up to 55, 39, and 29 %, respectively, compared to the high-performing information retrieval methods used in feature location. Moreover, the proposed time-aware technique increases the accuracy, performance, and effectiveness of the typical term-weighting technique, tf-idf, as much as 15, 11, and 13 %, respectively. Finally, the proposed approach outperforms our previous approach, noun-based feature location, as much as 17 %. These experimental results demonstrate that time-aware analysis of developers’ expertise significantly improves the feature location process.