Software Defect Prediction Based On Social Software Engineering

As current software engineering research focus, social software engineering has been studied in-depth and a lot of achievements have been gained. Achievements are applied in many fields, such as expert finding, bug categorizing and the development of coordination pattern. Besides, software defect prediction plays an important role in improving software quality and balancing software costs. Their combination, which means predicting defects based on social software engineering should have great research and application value. However, this is just been proceeded and the existing research is mainly extracting features from the traditional social network for defect prediction while the method of building social network is simple and monotonous, there also exists a lack of analysis and research on the unstructured communication information.This paper explores the communication and collaboration information of social software engineering separately based on the developer mailing list and social network analysis, and then extracts new defect prediction features:1) Link bug tracking databases to the mailing list archives. Then perform experiments through three aspects including the structure of message, both positive and negative emotions and the topic of mailing list with different approaches. Based on the exploration, mailing list features that may be correlated with source code defects are extracted.2) Improve social network from the perspective of construction method. The combination of multiple data sources in software repository enables the new network to contain more information about file dependency, developer contribution and collaboration. Through exploration, several features such as degree centrality and betweenness centrality which are of strong correlations with the software defects have been extracted. Next social network is expanded for researching the status and influence of developers as well as the organization structure of projects. Then new defect prediction features are extracted as core developer workflow in organization.On this basis, this paper makes use of new features including mail content feature, mail network feature, emotion feature, topic-based feature, network centrality feature, core developer feature and organization structure feature based on social software engineering to conduct defect prediction. Then correlation experiment and contrastive analysis experiment are conducted with eclipse project as the dataset. Results show that features based on mailing list and social network analysis are correlated with defects between 0.389 and 0.473 and the precision and recall of new features are 83.1% and 81.2% respectively, which are better than existing social features. Besides, by combining social features into the state-of-the-art features, the effectiveness of the prediction model is obviously improved.