Enhancements for duplication detection in bug reports with manifold correlation features

Abstract

In software maintenance activities, bug report processing is a major task in deriving crucial information for bug fixing. Because a considerable fraction of bug reports comprises duplicates in many projects, the duplicate reports must be identified for processing efficiency. Various text mining schemes have been proposed to handle this detection problem. This paper proposes an enhanced support vector machines (SVM) model (SVM-SBCTC) by considering the manifold textual and semantic correlation features based on a previous SVM-based discriminative scheme (SVM-54). We conducted empirical studies on three open source software projects: Apache, ArgoUML, and SVN. Compared with the SVM-54 scheme, SVM-SBCTC demonstrates promising detection performance in achieving relative improvements ranging 2.79%–28.97% in the top-5 recall rates among three projects. Furthermore, SVM-SBCTC demonstrates the top performance among various other weighting schemes in most cases.