Research On The Quality Assurance Of Crowdsourced Software Test

Crowdsourced software testing is a new paradigm of distributed testing and production organization,which can obtain abundant real users' feedbacks quickly and less costly.Many techniques have recently emerged to apply crowdsourcing to certain types of software testing such as Web application testing,open-source testing,and game beta testing.While the benefits of harnessing the collective efforts of individuals are obvious,less is known about the quality assurance of testing in a crowsourced environment.So an interesting and challenging topic is to consider how to integrate the mass and computing resources on the Internet to test collaboratively and to yield quality testing outcomes.This thesis aims at improving the quality control of crowsourced software testing,especially in terms of task assignment and bug report processing.And the contributions are as follows:(1)We propose a cooperative framework for crowdsourced testing based on mulitiple task matching(PM2CT),while improving test quality at the phase of task partition,assignment and result aggregation.When applying to Web application testing,multi-task matching algorithm can adatively assign certain tasks to appropriate testers by skill matching heuristics to test complicate pages more in order to ensure test coverage and reliability of testing results.Experimental results show that PM2CT is comparatively better than CPLEX tools,and also outperforms a representative crowd-based Web application testing technique..(2)To aggregate noisy and imbalanced bug reports,we propose a method FER for identifying bug severity by using an instance fuzzy entropy and RSOMTE.FER features the fuzzy k-nearest neighbor rule(F-KNN)for removing less important bug reports,and an improved Synthetic Minority Over-sampling approach(RSMOTE)for curbing the distributionimbalance.Conducted on real bug reports from three open source projects,several experiments statistically indicate that the present approach is robust against real imbalanced data while predicting the severity of bug reports effectively.(3)We propose to use feature selection and RSOMTE to identify high-impact bug reports.To do so,we first employ three feature selection algorithms to remove noisy and less-discriminative features,RSMOTE is next applied to enhance training sets.Several experiments have been conducted on datasets Camel and Wicket obtained from real-world bug repositories.The results show that the proposed approach can tolerate real imbalanced data while effectively identifying the high-impact bug reports.(4)Focused on the uncertainty arised from the random sampling by RSMOTE,we propose an approach FC-FI to predicting bug severity by integrating multiple RSOMTE-enhanced classifiers with Chouqet fuzzy integral.To do so,we train the classifiers respectively with the balanced datasets obtained from RSMOTE.Then,we use fuzzy integral to integrate them to obtain the ultimate prediction results.Several experiments have been conducted on bug reports from ten datasets of three large open source projects,namely Eclipse,Mozilla and GNOME.The results show that our approach can effectively reduce the data scale and improve the performance of identify the severity of bug reports.(5)To tackle the insufficient annotated data problem in some platform,we propose a Knowledge Transfer Classification(KTC)approach based on text mining and rough set.KTC uses an Importance Degree Reduction(IDR)strategy based on rough set to extract characteristic keywords when training on annotated bug repositories,and transfer what learned to predict the severity of unannotated Android bug reports.The results of several experiments indicate that KTC approach is beneficial for predicting the severity of Android bug reports.