The Research Of Crowdsourcing Test Report Prioritization Technology For Mobile Applications

Software testing is an indispensable part of software engineering,and crowdsourcing testing is an important branch of software testing.In crowdsourced testing,the workers conduct testing tasks and submit test reports,and developers need to review and evaluate the submitted test reports.Due to a large number of test reports and the uneven quality,developers will spend a lot of time in the manual review process,which directly affects the efficiency of crowdsourcing testing.In recent years,many automated techniques,such as clustering,classification,and prioritization techniques,have emerged to reduce the number of reviews and improve review efficiency.However,when faced with the text and image information of mobile crowdsourced test reports,existing report analysis techniques do not deeply explore the relationship between text and image information.Given the above situation,starting from the characteristics of the report itself,this thesis studies how to maximize the use of text and image information,and proposes three improved mobile crowdsourced test report prioritization techniques.ⅰ.To solve the problems of short text description and insufficient information in mobile application test reports,this thesis presents a test report prioritization technique based on data enhancement.This technique firstly segments the text information in the report,and uses the OCR technology to extract the keywords in the screenshots in the report.When these reports have fewer keywords,we summarize the keywords and generate new keywords with the help of similar reports.Finally,according to the keyword set of the report,we sample from these keyword sets by the greedy strategy algorithm,which can improve the efficiency of inspecting reports.ⅱ.Considering that existing prioritization techniques only segment text information,lacking deep analysis and understanding of the text content,we propose a test report prioritization technique based on text and image understanding.This technique classifies text descriptions into two categories,describing system behavior and reproduction steps,and we use different text feature extraction methods for these two categories.Then,the text features and the reduced dimensionality image features are joined together,and the inspection order is generated by sampling the clustered report clusters.ⅲ.Existing prioritization techniques only focus on the efficiency of reviewing different bugs,and do not consider the bug severity level,so a crowdsourced test report prioritization technique that considers the bug severity is proposed.This technique extracts features from the text and screenshot information of the test reports,uses the hash technique to index the test report,then samples the data in the hash table,and selects a report with the largest text information entropy in the data corresponding to each index,so as to obtain the inspection order.Meanwhile,in these two modalities of information,the image conveys more objective information,so we make more use of image features.Aiming at the characteristics of short text descriptions and rich screenshots of crowdsourced test reports in mobile application scenarios,this thesis proposes three prioritization techniques,which can reduce the number of reports reviewed manually and effectively improve the efficiency of inspecting reports.