The Evaluation For Priority Of Developers And Bug Reports In Crowdsourcing Test Platform

At present,many large software projects use crowdsourcing test platforms to collect and manage the test reports which are submitted by software testers.The difference between performance of testers in submitting and commenting to bug reports reflects their contributions to software testing is different.And the bug reports submitted by different software testers also have different influence on the Software system maintenance.Thus,evaluating the testers' contributions on crowdsourcing test platform and identifying high-impact test reports reasonably can increase the testers' enthusiasm and improve test efficiency.However,with the repaid increasement of the bug reports in software projects,accurate evaluating the testers' contributions to the projects becomes more and more difficult.To solve the above problems,we regard the submitting and commenting to the bug reports as social behaviors of the testers,and we establish a network of testers based on the tester's work behavior.Then,we proposes a ranking method to evaluate the tester's contribution to software testing.The case study of the Eclipse project on Bugzilla shows that the result produced by the approach ranking the influence of testers is consistent with the actual performance of the testers in this project,and the result is also stable and effective.Therefore,it is proved that the approach we propose can be reasonably used for measuring the contributions of the testers and providing basis of reward distribution for software testers.Simultaneously,due to the large number of test reports,the developers do not have enough time to inspect all the reports.Thus,they often focus on inspecting and repairing the highly impacting bugs.Some methods have been proposed in recent years,but the performance is still not satisfied.In order to improve the performance mentioned above,an integration method based on machine learning to assist developers predict high-impact bugs is presented in this paper.Firstly,we use weight to optimize the discriminant probability of each classifier output according to the data distribution of different items and classification characteristics of classifiers.Then,several optimized classification algorithms are integrated to identify test reports with high impact in unbalanced data.The experiments show that the base classification algorithms used in this method are reasonable and the method we propose performs better in terms of detecting high-impact bugs than previous methods.