Towards the identification of bug entities and relations in bug reports

摘要

During the bug fixing process, developers usually analyze the historical relevant bug reports in bug repository to support various bug analysis and fixing activities. There are rich semantics and relationships in the bug reports, which can be helpful for bug retrieval, recommendation, and repair. In this paper, our purpose is to quickly extract effective knowledge of bug report from two perspectives: entity recognition and relation extraction to assist bug understanding and fixing. Meanwhile, we hope to strengthen the relevance of bug reports through the effective extraction of bug knowledge. In order to effectively extract the bug entities and relations in the bug report, we first define 8 types of relations between the bug entities and incorporate neural network Recurrent Neural Network (RNN) and RNN based on shortest dependency path (SDP-RNN) to automatically identify bug entities and their relations in bug reports. Results We evaluate the effectiveness of our method through four experimental questions. From the results, the bug knowledge extracted by our method can effectively represent the semantics and relations in the bug report, and obtain F1 scores of 79.32% and 63.8% in entity recognition and relation extraction, respectively. The proposed approach can efficiently extract the structured bug knowledge in the bug report, and further enhance the correlation between the bug reports and the effectiveness of the bug knowledge through the representation of these structured bug knowledge units.