An Improved Discriminative Model for Duplication Detection on Bug Reports with Cluster Weighting

Abstract

Processing bug reports plays an important role for software maintenance. Recently, the issue of detecting duplicate bug reports has been noticed due to their considerable appearances. In the past, many NLP-based detection schemes have been proposed. However, the cluster-level correlation relationships are not extensively considered in the past studies. In this paper, we present an improved detection scheme using cluster weighting to enhance the detection performance of a previous SVM-based method. We have conducted empirical studies with three open source software projects, Apache, ArgoUML, and SVN. Compared with the original SVM-based method, the proposed SVM-TC scheme can achieve 2.83-16.32% improvements of the top-5 recall rates in three projects.