Research On Spatial Crowdsourcing Task Assignment Based On Recommendation

With the full popularity and coverage of mobile devices and wireless networks,the concept of Spatial Crowdsourcing(SC)emerged.The spatial crowdsourcing system consists of crowdsourcing workers,crowdsourcing tasks,task publishers,and spatial crowdsourcing platforms.In the spatial crowdsourcing system,workers carrying portable devices need to move to the task location specified by the platform to perform tasks.Task assignment is a key problem in the field of spatial crowdsourcing.Task assignment requires spatial crowdsourcing platforms to assign/recommend suitable tasks for workers under the spatio-temporal constraints of workers and tasks.In different scenarios,workers have different preferences for different tasks.In order to learn workers’ various preferences for tasks,this thesis studies the recommendation-based spatial crowdsourcing task allocation technology,and obtains the following research results:1)This thesis combines point-of-interest recommendation with task assignment,studies the recommendation-based spatial crowdsourcing task assignment technology,formally defines the task assignment problem based on workers’ spatio-temporal preference and the task assignment problem based on worker’s sequence preference,and puts forward respectively Corresponding technical frameworks.2)Aiming at the problem of task assignment based on spatio-temporal preference,this thesis proposes a data-driven task assignment technology framework based on worker spatio-temporal preference,which is divided into spatio-temporal preference learning stage and task assignment stage.In the spatio-temporal preference learning stage,this thesis adopts a recommendation model based on relational translation to predict workers’ spatiotemporal preference scores and jointly model workers’ temporal and spatial dynamic preferences.In the task allocation stage,considering workers’ spatio-temporal preference scores,this thesis designs a greedy and Kuhn-Munkras(KM)based task allocation algorithm to achieve optimal allocation.3)In the task assignment problem based on worker’s sequential preference,in order to learn the worker’s sequential preference,this thesis adopts graph convolutional neural network model and recurrent neural network model for joint learning.In the task allocation part,this thesis transforms the task allocation problem based on sequence preference into the minimum cost maximum flow problem,and designs two task allocation algorithms.4)This thesis conducts extensive experiments through real check-in data sets.The experimental results verify the effectiveness of the two technical frameworks proposed in this thesis,and can better solve the two new problems raised in thesis.