Research On Quality Control Issues Based On Social Networks In Crowdsourcing

With the continuous development of Web 2.0 technology,crowdsourcing has become an important way to share knowledge,collect information,and complete complex tasks.More and more workers are participating in the crowdsourcing system,which leads to the fact that crowd workers have different social backgrounds,ability levels and motivations to participate,and the quality of the completed tasks also varies greatly.Therefore,how to ensure the quality of crowdsourcing tasks has become a hot research issue in the current crowdsourcing field.Many current studies assume that workers work independently,but it turns out that crowd workers will cooperate in teams when performing tasks,and even collusion each other up for less labor for task bonuses,forming collusion similar to the ’free-rider’ in economics.Collusion can easily lead to the same task results submitted by lots of workers,and will affect the motivation of independent workers,thereby reducing the quality of crowdsourced data.In view of the above problems,this dissertation carries out two aspects of work to control the quality of crowdsourcing.(1)This dissertation accurately demarcates the boundaries of collusion groups in crowdsourced social networks to detect collusion workers.Then measured the interaction strength of workers in a crowdsourcing social network and inferred the relationship between crowdsourced answer repetition rate and worker intimacy.This dissertation proposed a new calculation method of worker intimacy,and a collusion team detection algorithm is proposed based on the worker intimacy: Trust-Team.The algorithm can quickly identify collusion teams in crowdsourced social networks and set a penalty function according to the size of the collusion team and the closeness of workers.Finally,experiments are carried out on Yantai University’s teaching evaluation dataset and three public crowdsourcing datasets,it has been proved that the algorithm can effectively improve the platform’s efficiency to identify collusion behaviors and improve the accuracy of crowdsourcing results after selectively filtering the answers of collusion workers.(2)This dissertation proposes a crowdsourced result aggregation algorithm based on workers’ credibility for processing crowdsourced data after removing collusion answers.First,an elastic credibility model considering collusion factors is established to evaluate the credibility of crowd workers.Then workers’ credibility is used as a weight to introduce an improved DS model,which is called the Trust-FDS algorithm.Finally,this dissertation verifies the algorithm on the teaching evaluation dataset of Yantai University and two public crowdsourcing datasets.The experimental results show that the Trust-FDS algorithm can converge faster while maintaining the correct rate,improving crowdsourced data quality.