Research On Task Prediction And Assignment In Spatial Crowdsourcing

With the development of wireless networks and the popularization of intelligent mobile devices,spatial crowdsourcing has been integrated into human life from all aspects of food,clothing,housing,and transportation.Task assignment in spatial crowdsourcing refers to the assignment of multiple tasks to multiple workers on the spatial crowdsourcing platform under the given constraints.To simplify the algorithm design,most of the current task as-signment methods assume that the tasks and workers on the platform are fixed.However,in practical application,this assumption is too simple,both workers and tasks are changing dynamically in real-time.If we can capture this dynamic and apply it to task assignment,it will form a triple-win situation:workers can arrive at the task site in advance to com-plete more tasks;tasks of task publishers will be completed faster,so as to have a better user experience;the platform can effectively configure workers' resources and improve the total revenue of the platform.In view of the above shortcomings of the current work,this paper mainly studies the task prediction method in spatial crowdsourcing and how to use the predicted data task assignment.The specific works are as follows:1.It is very difficult to predict the release time and location of a single task accurately,to predict the total number of tasks in a specific region and specific time period is firstly considered.This paper proposes a sequential spatio-temporal residual neural network to analyze and mine historical data.The network captures the temporal dependence of different time granularity by serializing the historical data of different granularity and uses the characteristics of the convolutional neural network to capture the spatial dependence.Experiments on real datasets verify the validity of the model.For regions with intensive task requests,the prediction effect of the model is more significant.In order to reduce the influence of abnormal data on the model,the input sequence is smoothed,which improves the accuracy by about 7%.2.External information(such as weather,holidays,and other data)is often used to im-prove the accuracy of the model.However,the external information is not only noisy and difficult to collect,but also very complex and difficult to make full use of in the model.To solve this problem,the original spatio-temporal data is used to capture ex-ternal information implicitly,and three attention mechanisms are designed to increase the positive effect of strong correlation information and reduce the negative effect of weak correlation information.The validity of the model is verified on the real data set,and the model can effectively capture external information from the weather aspect.3.In order to increase the benefits of workers,task publishers,and the platform,this paper proposes a downwind task assignment mechanism based on the results of task prediction.This mechanism designs a route planning scheme to improve the number of task orders received by workers and sets the task completion efficiency ratio to ensure the user experience of task publishers.Experiments prove the effectiveness of the task assignment mechanism.