Research On Key Technologies Of Automatic Classification And Assignment Of Software Bug Reports

Classification and triage of bug reports is one of the main activities in the maintenance phase of software engineering,which has a significant impact on software quality assurance.The traditional classification and triage of bug reports is often done by experienced senior developers.This manual triage method is not only time-consuming,but also has a low accuracy rate.In recent years,the automatic classification and triage method based on machine learning has the potential to replace manual work.However,the existing automatic method uses a single report information and the extracted features are not significant,so it is difficult to accurately classify and triage bug reports automatically.This study focuses on bug reports of open source software,proposes a new automated method applicable to software bug report classification and triage,and finally improves the accuracy of automatic bug report classification and triage.The paper mainly conducts research from the following three aspects:(1)In view of the fact that the existing methods only use the report summary and description fields,and the text representation ability is insufficient,a bug report error classification method based on multivariate text mining and random forest is proposed.This method trains a random forest classifier by extracting text features,frequency features and the report intention features proposed in this paper from multiple metadata.Experiments are conducted with real bug report data in Bugzilla,and the results show that after adding the proposed report intention feature,the accuracy of classification increases by an average of 11%.At the same time,the F-measure of the proposed method combined with five different machine learning classifiers can reach 83% to 96%.Experiments verify the effectiveness and accuracy of the proposed method in solving the problem of classification of bug report error types,and also prove that the method has good scalability.(2)In view of the fact that the existing methods only consider the shallow text information of the report content and ignore its deep hidden emotional characteristics,a bug report severity classification method based on the emotional score optimization artificial neural network is proposed.The innovation of this study is to consider the relationship between the emotional characteristics and severity of the report summary text,and generate an optimization function by combining the emotional score with the output of the neural network model and the actual severity of the report through multiple linear fitting,and use the function to optimize the neural network.Compared with the benchmark method,the results show that the accuracy of the classification model after adding the optimization function is 3.3% higher than that of the ordinary artificial neural network model.Experiments verify the effectiveness and accuracy of the proposed method.(3)Aiming at the fact that the existing method only considers the text content of the report itself when bug triage,and ignores the features of the developer side,resulting in low dispatching accuracy,a bug report triage method based on multi-feature fusion is proposed.The method firstly extracts the reported text sequence and local features and fuses the error type and severity features to train an automatic assignment model.Then,the relationship between the developer and the report type is mined to evaluate the developer,and the ranking of the developer recommendation output by the assignment model is optimized by using the developer evaluation score.Through the experimental comparison with the baseline method,the results show that the accuracy of the proposed method is 2.1%,3.7%,and 3.1% higher than the baseline method when predicting 1,3,and 5 developers,respectively.Experiments have verified that it is effective and feasible to apply the proposed method to the actual bug report classification and triage tasks.