Analysis And Research On The Crowdsourced Team Formation

In recent years,crowdsourcing has greatly promoted the development of software engineering activities,such as crowdsourcing software development,crowdsourcing testing,etc.,because of its high quality and low cost.However,the total number of tasks in crowdsourcing platform is increasing day by day,which makes it difficult for crowdsourcing workers to select tasks and task publishers to select candidates.Task publishers need to select high-quality workers from hundreds of unfamiliar candidates to form teams to accomplish tasks on the premise of controlling budget and team size.As crowdsourcing virtual team is an important part of successful crowdsourcing task,previous studies have tried to use many algorithms to solve the problem of crowdsourcing team formation,including Genetic Algorithm(GA),Alternating Variable Method(AVM),but the performance is still limited.Considering the influence of worker's ability and salary on successful software tasks,we present the mathematical definition of crowdsourcing team formation problem.This problem is a typical discrete combinatorial problem with great complexity.In this paper,two intelligent optimization search algorithms CVTMaker and ESSA-rank are proposed to help task publishers construct ideal virtual crowdsourcing teams.CVTMaker uses dynamic real number to encode team solutions.To overcome the shortcoming that(1+1)-ES is easy to oscillate near the global optimal solution,AVM is used to search the local space of the current best individual.The two parts are executed alternately to coordinate the exploration and exploitation ability of the algorithm,thereby improving the overall performance of the algorithm.Compared with CVTMaker,ESSA-rank can search more individuals in each iteration.The discrete reorganization operator with fitness-based ranking selection is applied to increase the probability of high cost-effective workers being selected in team solution,and the algorithm is guided to converge gradually to the global optimal solution by the parameter adaptive mechanism.In this paper,we construct a data set containing 1556 instances of close-to-real crowdsourcing team behavior by using knowledge transfer method to mine Q&A information from the Stack Overflow of Q&A community.Then experiments are conducted to evaluate the performance and scalability of CVTMaker and ESSA-rank algorithms on classical dataset and constructed dataset.The results show that CVTMaker and ESSA-rank can search better team solutions with a large extent of superiority than GA and AVM in both two datasets.When dealing with crowdsourcing team formation with different team sizes,CVTMaker has more advantages in dealing with problem with large team sizes,while ESSA-rank is more suitable for dealing with problem with small team sizes.