Software Defect Classification Approach Based on the Modified Latent Dirichlet Allocation Topic Model Considering the Domain Characters

Abstract

The existing defect classification approaches do not consider the domain characters of software defects, such as aeronautics domain, astronautics domain and so on. Therefore, the precision rate and the recall rate of these defect classification approaches are not very accurate in many conditions. To resolve this problem we present a new defect classification approach based on the modified Latent Dirichlet Allocation (LDA for short) topic model combining with domain characters to improve the performance of the defect classification. First, we propose the defect segmentation approach based on the special domain characters. Then, we propose a modified LDA topic model combining with software requirements. Based on the proposed modified LDA model, we obtain a new defect classification approach. Finally, the experiment result shows that the precision rate and the recall rate of the defect classification are au improved up to 15%~20% compared with the existing classification models. Thus, we consider that this new classification approach is very suitable for classifying the defects with obvious domain characters.