Study On Task Assignment In Spatial Crowdsourcing Based On Collaborative Filtering

With the rapid development of mobile intelligent terminal equipment in recent years,more and more spatial crowdsourcing platforms have begun to emerge,such as Didi taxis,Meituan meal ordering service,etc.Spatial crowdsourcing has begun to slowly integrate into our daily life,and closely related with our life.With the development of spatial crowdsourcing,the kind of spatial crowdsourcing has become not only limited to standardized tasks such as call tax and delivery,but also the emergence of some non-uniform standard tasks such as haircut,etc.In order to complete the spatial crowdsourcing,crowd workers need to move to the designated location to complete the task,which will result in travel costs.Smaller travel costs mean less response time and higher task acceptance rate.At the same time,in view of non-uniform standardization of tasks,different task requesters are dissatisfied with the satisfaction of the same crowd worker to accomplish their tasks.It is important to assign the satisfaction of crowd workers to the task requester.This will enhance the task requester's dependence on the spatial crowdsourcing platform.Therefore,this paper proposes a collaborative filtering-based task assignment in spatial crowdsourcing,in order to obtain a smaller global travel costs and assign the appropriate worker to task requester.The work done in this paper is as follows:(1)In this paper,a prediction-based method is proposed to predict the number of task at different task locations to optimize the overall minimum travel cost assignment problem.Because workers need to arrive at a designated place to complete the spatial task,the task assignment result in this time instance will affect the task assignment in the next time instance.However,as the location of a spatial task in the next time instance cannot be predicted.In this paper,we propose a prediction-based method to optimize the overall minimum travel cost by using the local minimum travel costmaximum expected number of tasks.(2)In the prediction of the number of crowdsourcing tasks in the next time instance,this paper proposes a crowdsourcing task number prediction method based on Bayesian classification,regression-based and Laplacian regularization.Based on the grid,the selected area is divided into grids of the same size,and the number of historical tasks in each grid is used to predict the number of crowdsourcing tasks.(3)This paper combined with the user experience of requester,combines the method of collaborative filtering-based and the method of prediction-based to assist in task assignment.We use memory-based collaborative filtering and model-based collaborative filtering to predict the scoring matrix of the crowd workers by the task requesters respectively,combined with the distance matrix of task requesters and crowd workers to obtain the mass travel cost matrix of task requesters and crowd workers.We use the local minimum unit mass travel cost-maximum the expected number of tasks to optimize overall minimum unit mass travel cost,to assigning the appropriate crowd workers to assign requesters with the lowest possible travel costs,thereby enhancing the assign requester's dependence on the platform.(4)In this paper,g Mission and Yelp datasets are used respectively to conduct experiments,the experiments show that this method can effectively reduce the overall travel cost and the overall unit mass travel cost.