A Similarity Integration Method based Information Retrieval and Word Embedding in Bug Localization
Shasha Cheng, Xuefeng Yan, Arif Ali Khan · 2020
To improve the performance of bug localization, there is necessity to solve the lexical mismatch between the natural language in the bug report and the programming language in the source file. A similarity integration method for bug localization is proposed, in which the similarity between bug report and source file is calculated by information retrieval (IR) and word embedding. More specifically, IR technique is used to collect the exact matches between bug report and source file. The terms in the bug report and the potential source files of different code tokens are connected by word embedding technique, which is used to complement with IR technique. Finally, deep neural network (DNN) is utilized to integrate extracted features to get the correlation between bug reports and source files. The experimental results show that the proposed approach outperforms several existing bug localization approaches in terms of Top N Rank, MAP, and MRR.