| In the process of software development and maintenance,Bug is almost an unavoidable problem for all developers and testers.Bug tracking management system is a kind of software that helps developers and testers maintain bugs in their work,and is used to manage and track bugs in the system.The bug tracking system contains a large number of bug reports.Developers usually use bug tracking management systems or search engines to assist in the analysis and fix of current new bugs.However,there are many source code files involved in bug reports in the bug tracking management system.If the developer has been informed of the similar bug reports of the current bug in advance,the developer can focus on fewer source code files,thereby reducing the workload of program understanding.In addition,it is difficult for search engines to accurately understand the search intent of developers,and thus cannot provide accurate information to assist in the analysis and fix of current bugs.To alleviate the above-mentioned problems,this thesis constructs a bug knowledge graph and recommends similar bug reports for developers based on the bug knowledge graph.At the same time,based on the bug knowledge graph,we construct an intelligent Q&A to accurately answer bug-related questions,thus improving the efficiency of fixing bugs.The work of this thesis mainly includes the following aspects:1)In view of the large number of similar bug reports contained in the bug tracking management system,we propose a knowledge-driven method for recommending similar bug reports.First,we analyze the historical bug reports and mine the related information of the bug and the relationship between the bugs,and construct the bug knowledge graph.On the basis of the bug knowledge graph,we combine information retrieval with Word Embedding,and we take various attributes of the bug into consideration.Finally,we conduct comparative experiments with the baseline method on the Firefox dataset.The experimental results show that the proposed approach improves the performance by an average of 12.7%on the Firefox dataset compared to the baseline method.2)To alleviate the problem of inaccurate and incomplete information provided by search engines,we propose a Q&A method based on pre-training model and knowledge graph.On the basis of successfully constructing the bug knowledge graph,the pre-training model is applied to the whole process of bug knowledge Q&A,such as entity recognition,entity linking,answer sorting and other modules.In order to evaluate this method,a comparative experiment was carried out on the Mozilla dataset with the commonly used methods of knowledge Q&A.From the experimental results,the accuracy of this method is improved by 11.02%compared with Bi-LSTM method and is improved by 8.05%compared with Bi-LSTM-CNN method.3)We construct a bug data intelligent analysis platform based on the bug knowledge graph.The platform is composed of modules such as entity query,relationship query,similar bug recommendation,and bug knowledge Q&A.The bug data intelligent analysis platform is mainly used to integrate and manage bug reports and committed information helping developers understand and fix bugs. |
PDF Full Text Request