The Research Of Crowdsourcing Test Report Analysis Technology Based On Image Understanding

Crowdsourced testing has been widely used to improve the software quality of mobile applications due to its rapid iteration,low cost,and real feedback.A crowdsourced worker performs tasks on a crowdsourced platform and presents experiences in the form of test reports,inspecting a large number of reports with varying quality in a limited time becomes a time-consuming yet inevitable task.In recent years,many text-based analysis techniques have been proposed to overcome this challenge.However,in mobile testing,test reports usually contain only short text descriptions but contain rich screenshots,which makes the application of existing report analysis technology become a barrier.Compared with the uncertainty of textual information,the screenshots can adequately represent the usage scenarios of mobile applications and objectively reflect the design defects in the software system.Because of the above situation,this paper has optimized and innovated the traditional report analysis technology,and mainly proposed three methods to assist developers in analyzing and understanding the test report:i.To alleviate the redundancy of crowdsourced test reports,this paper proposes a clustering sampling technique of test reports based on image understanding.This technique processes the text and image information respectively,then uses similarity measurement and balance algorithm to establish the distance matrix of the test report.By sampling the report cluster after matrix clustering,the defect types in the report set can be detected efficiently.ii.To solve the problems of text information shortage and image reading difficulty in the crowdsourced test report,this paper proposes a text generation technique for test report screenshots.Firstly,images with similar content will be clustered,and a probabilistic language model is built to generate keyword sequences for each image cluster.When a new report is to be inspected,a precise matching mechanism is used to generate a corresponding text description for the screenshots in this report.iii.Considering the short text description and rich image information of crowdsourced test report,this paper proposes an automatic classification technique for test report based on image understanding.This technique extracts textual features and image features at the same time,then combines the idea of feature fusion with the existing classification algorithm.On the premise of effectively improving the classification effect,this paper also discusses the most suitable classification model and training set size for crowdsourced test report.