Research On Crowdsourcing Task Assignment By Worker Interest

Crowdsourcing is a new distributed problem solution.It uses the Internet to gather the wisdom or skills of people to solve problems that machines cannot handle,such as image annotation,entity resolution,and creative collection.With the widespread application of crowdsourcing,task assignment has attracted the attention of many researchers.Most existing studies about crowdsourcing tasks assignment are based on task quality,economic cost and completion time,rarely considering the worker's own interest.It happens that workers refuse to participate in tasks assigned,and the completion rate of task is reduced.The traditional methods considering workers' interests,such as collaborative filtering algorithm,use scoring data to calculate the similarity of workers or tasks,which are faced with problems such as sparse scoring data.In addition,the existing methods have not considered the influence of time factors on workers' interests.Assuming that workers' interests are fixed,crowdsourcing platform will assign same type of tasks to workers,which make workers boring.To solve the above problems,we study the crowdsourcing task assignment algorithm based on worker interest.Our research work is as follows:(1)A task assignment method based on domain interest is proposed.First,we propose a collaborative filtering algorithm based on worker domain and rating(WDR-CF),which constructs workers' domain interest model through the rating frequency and domain label scores,and uses the modified domain interest model to improve the calculation of worker similarity.Then,a collaborative filtering algorithm based on task domain and rating(TDR-CF)is proposed.The correlation between tasks is calculated through a combination of task domain and task rating,which improves the calculation method of task similarity.Finally,a hybrid collaborative filtering algorithm(DR-HCF)based on domain and rating is proposed.The prediction scores of unknown tasks by WDR-CF and TDR-CF algorithms are combined by linear weighting,which improves the original prediction score method.Experimental results show that compared with existing algorithms,our algorithm is better in scoring prediction accuracy on the data set with sparse score data.(2)A time-sensitive task assignment method is proposed.The time weight for the domain label is defined to quantify the workers' interest in the domain that changes with time,so as to construct a time-sensitive worker interest model.We integrate this model into the collaborative filtering algorithm,and propose a time-sensitive collaborative filtering algorithm(TS-CF),which uses time-weighted domain interest to improve the calculation method of worker similarity.The experimental results show that the accuracy rate has been significantly improved by introducing the time factor.Furhermore,based on the matrix factorization algorithm,we propose a matrix decomposition model by worker similarity(WSBSVD++),which uses neighboring workers' ratings to improve the original score prediction method.The experimental results show that the accuracy of the WSBSVD++ model significantly outperforms other models.