The Research On Automatic Labeling Method For Mobile Application Issue Reports

In recent years,with the popularity of mobile devices such as smartphones and tablets,the number of applications running on mobile devices has increased dramatically.At present,there are millions of mobile applications in the mobile application market.The maintenance of mobile applications of this scale is a huge challenge for developers.Developers need to maintain these applications,so,they can continue to benefit end users.They usually submit issue reports to describe bug,feature requests,and other changes that occur in the app.Labels(e.g.,bug,feature requests,etc.)are important attributes that indicate which issue reports should be resolved first or next.If the label of the issue report is "bug",this means that the report describes a serious error and the developer should fix the error first.Otherwise,if the label is not a ‘bug’(for example,‘feature request’,etc.),the developer can fix it later.Studies in recent years have shown that more than 30% of issue reports in issue report repositories are not labeled.For these unlabeled issue reports,developers need to spend extra time to manually verify each unlabeled issue report so they can decide to resolve the most important issues.Since manual verification of each unlabeled issue report requires a lot of human resources and time resources,it is wise to develop an automated method for the machine to automatically label unlabeled issue reports.To this end,this paper proposes a new method for automatically labeling unlabeled issue reports.The main research contents of this paper are summarized as follows:This paper combines the term-frequency-inverse-document-frequency method,cosine similarity(cosine similarity)calculation method,word2 vec,Microsoft Concept Graph,the detection method BM25 Fext and Jaccard similarity coefficient applied to the duplicate report in the issue report repository,the SURF method which classifies user reviews,and designs four text similarity measurement methods to calculate the text similarity between each unlabeled issue report and the labeled reports,as well as the text similarity calculation between each unlabeled issue report and user reviews,If the hybrid similarity score is higher than the threshold,the issue report is labeled as “bug” or “feature request”.Otherwise,the issue report will be labeled as “other”,which means it will not describe a software bug or feature request.Finally,this paper conducts an experimental verification study on the proposed new method.In order to evaluate the effectiveness of the method,we conduct the experiments in17 open source mobile apps.The method was tested based on these data.In this paper,F1 evaluation matrix is used to express the performance of the method.By comparing with the previous research methods,it is found that the F1 score of the method proposed in this paper is higher than that of the previous research methods,which shows that the method proposed in this paper can accurately label the unlabeled issue report and achieve high performance.