Improving Detection Performance of Duplicate Bug Reports Using Extended Centroid Features

Abstract: According to recent work, detection on duplicate bug reports has received much attention. One of the reasons is that duplicate bug reports may consume time of bug triagers and software developers. In previous studies, many schemes have been developed for using text mining techniques or using the information retrieval and natural language processing techniques. In this paper, we propose a method to improve centroid characteristics by adjusting centroids with better initial values than based on Class-Feature-Centroid (CFC) [12]. With the effectiveness of CFC, the centroid- based approach can obtain further improvements for detection performance. The method includes two steps. First, we extract inter-class and inner-class term indices from the corpus. Second, we enhance centroid calculation based on class features. Moreover, for similarity measure we also adapt the calculation of the traditional cosine similarity by denormalized cosine measure which is also used in [12].

Keywords: Bug Reports, Duplication Detection, Feature Weighting, Class-Feature-Centroid