Research And Application Of Quality Control In Crowdsourcing

With the rapid development of Internet technology,more and more smart product users are participating in crowdsourcing systems.In particular,the derivative and rapid development of the sharing economy has made the crowdsourcing system and life closely integrated,and various crowdsourcing platforms have emerged endlessly,such as Baidu Know,Yahoo Answers,Meituan Takeaway,Didi Taxi,etc.Crowdsourcing includes traditional crowdsourcing and space-time crowdsourcing.The former is mainly online,and the latter is a combination of network and space activities.No matter what kind of crowdsourcing system,workers need to feedback the results of completed tasks to the crowdsourcing platform.Due to the free and loose working method of the crowdsourcing system,the crowdsourcing workers have acts such as deception and collusion in completing the task.In order to control the completion quality of crowdsourcing tasks,mining low-quality crowdsourcing workers and reducing their impact has become one of the research hotspots in the field of crowdsourcing.In view of the above problems,this paper analyzes the advantages and disadvantages of the existing crowdsourcing quality control methods.Through evaluating and analyzing the output data of the task,a crowdsourcing quality control model is proposed,and experiments are performed to verify its effectiveness and accuracy.The main research contents of this article include the following aspects:(1)Aiming at objectively evaluating crowdsourcing tasks and workers' unified bullying behavior,a weight setting algorithm based on Generalized Pareto Distribution(GPD)is proposed.The algorithm performs maximum likelihood estimation on the Pareto function,obtains the corresponding likelihood function and uses the dichotomy method to approximate the zero point of the function,calculates the Pareto shape and scale parameters,and obtains a fitted function expression.Corresponding weight formulas are defined in the algorithm,and the algorithm is used to process the task output data of crowdsourced workers,and an influence weight can be set for it.(2)For the purpose of subjective evaluation of crowdsourcing tasks and collusion among workers,this article first proposes two ways to divide the type of collusion,and excavate collusion groups on the premise of different types of collusion.Then use the structure and closeness of ties in the gang to infer the collusion relationship between them,and then calculate the intimacy value between the workers with the collusion relationship,and draw a collusion network diagram.Finally,use the obtained collusion network graph and the intimacy between the workers to calculate the reputation value of each collusion worker.