Research On Software Bug Report Recommendation Method Based On Feature Importance

Since the development and maintenance of open source software adopts voluntary participation and decentralized organization,compared with commercial software,open source software bug dispatch has more uncertainty and often has a longer bug repair process.Some large-scale open source software such as Mozilla and Eclipse have a wide range of users.With the iterative upgrade of the version,the functions are becoming more and more complex,and a large number of bug reports are generated every day.Due to the uneven level of bug submitters and experience levels,there are a large number of low-quality,repetitive or even invalid bug reports.These reports directly affect the efficiency of the entire bug dispatch and repair;on the other hand,defects require a lot of structured and semi-structured Information description,what information needs to be elaborated? What information can be briefly described? Clarifying these issues can not only effectively guide the submitter to submit high-quality bug reports,but also assist bug dispatchers and repairers to quickly focus on the key features of defects,form accurate determination of bug validity and accurate prediction of difficulty,and directly and indirectly improve Bug dispatch efficiency.Most of the traditional bug assignment recommendations are based on the text characteristics of defects and are implemented by various machine learning methods,but the accuracy of prediction has been low.This paper comprehensively considers various types of defects,analyzes the importance of different types of features,designs a deep learning model that can both complete the corresponding bug prediction and achieve feature importance measurement,and design fusion feature weights on this basis The similarity measurement method is combined with the historical bug dispatching graph(Tossing Graph)to optimize the recommendation list and improve the accuracy of bug dispatch.The main contributions of this article are as follows:1)Design a bug data crawling and information extraction program to complete the data crawling of the Mozilla open source project,and can realize automatic data extraction,data clarity and feature extraction to form an open source bug data set;2)Design a bug report description information structure that includes the publisher's experience,completeness,readability,social network and text in 5 dimensions and 28 features,and propose a deep learning model based on the Attention mechanism to achieve effective bug reporting Predict and repair fast and slow predictions,and can automatically measure the importance of features in the prediction process.The experimental results show that the model proposed in this paper is higher than the existing methods in validity judgment and bug repair time prediction,and can be combined with common feature selection methods to complete the feature importance ranking between and within dimensions;3)The developer prediction method BFIR model based on the feature importance of bug reports is proposed.The model recommends a list of developers who repair similar defects through the similarity matching method of bug reports based on feature importance,and is further optimized by the improved Tossing Graph method Recommended list.Experiments show that the method proposed in this paper is superior to common machine learning methods in bug dispatch recommendation.It also verifies that the multi-feature factors,feature importance factors,and improved Tossing Graph recommendation list optimization factors considered in this article can help improve bug dispatch Accuracy.