Mining And Processing Method Of Crowdsourced Test Reports By Fusing Text And Image Information

Crowdsourced testing is an emerging trend in the field of software testing,which has attracted widespread attention in the industry and academia.Crowdsourced testing is reliable,efficient,and fast,but the number of test reports submitted on crowdsourced platform is large and highly redundant.How to review high-quality test reports within a limited time is a challenge for report reviewers.In view of this,this paper proposes a clustering and ranking method for crowdsourced test reports,which integrates text and image information.(1)A clustering method for crowdsourced test reports that combines text and image information.Firstly,the text information and screenshot information in the test reports are respectively extracted,and are processed by the natural language processing technology and the spatial pyramid matching technology to calculate the text similarity and screenshot similarity between reports.Secondly,the weighted method is used to calculate the mixed similarity between test reports,based on which the test reports are clustered.Finally,the method is applied to the test report sets of five different applications.The experimental results show that the proposed method can improve the clustering accuracy of crowdsourced test reports.(2)A ranking method for crowdsourced test reports that combines text and image information.The similarity between test reports is calculated by processing text information and screenshot information.In addition,the defect detection degree of test reports is defined by combining text and screenshot information.Based on the defect detection degree and similarity of test reports,a two-stage sorting method is proposed to prioritize test reports.In the first stage,based on defect detection degree and similarity between reports,the test report sets are sorted and clustered,and the sorting results of partial reports and the similar test report sets of the sorted reports are obtained.In the second stage,the similar test report sets are sorted according to the criteria of minimizing similarity and maximizing defect detection.The sorting results of the two stages are merged to form the final test reports priority.Finally,experiments show that the proposed method can obtain high-quality test reports priority.The clustering and ranking method of crowdsourced test reports proposed in this paper can greatly reduce the cost of review by report auditors,reduce the cost of testing,improve the quality of testing,and thus contributes to building high-quality software.Therefore,this study has certain theoretical and practical value.