Detecting Crowdsourced Test Report Consistency for Mobile Apps with Deep Image Understanding and Text Analysis

Crowdsourced testing, as a distinct testing paradigm, has attracted much attention in software testing, especially in mobile application (app) testing field. Compared with in-house testing, crowdsourced testing outperforms because it utilize the diverse testing environments of different crowdworkers faced with the mobile testing fragmentation problem. However, crowdsourced testing also brings some problem. The crowdworkers involved are with different expertise, and they are not professional testers. Therefore, the reports they may submit are numerous and with uneven quality. App developers have to distinguish high-quality reports from low-quality ones to help the bug revealing and fixing. Some crowdworkers would submit inconsistent test reports, which means the textual descriptions are not focusing on the attached bug occurring screenshots. Such reports cause the waste on both time and human resources of app developing and testing. To solve such a problem, we propose ReCoDe in this paper, which is designed to detect the consistency of crowdsourced test reports via deep image-and-text fusion understanding. First, according to a pre-conducted survey, ReCoDe classifies the crowdsourced test reports into 10 categories, which covers the vast majority of reported problems in the test reports. Then, for each category of bugs, we have distinct processing models. The models have a deep fusion understanding on both image information and textual descriptions. We also have conducted an experiment to evaluate ReCoDe, and the results show the effectiveness of ReCoDe to detect consistency crowdsourced test reports.