| The rapid progress of mobile Internet technology has promoted the popularity and development of spatial crowdsourcing applications such as location-based service APPs and online cars.Spatial crowdsourcing refers to a new crowdsourcing computing model in which task performers(also called "crowdsourcing workers")complete crowdsourcing tasks with spatiotemporal characteristics issued by task requesters and meet constraints in an active or passive manner.At present,various spatial crowdsourcing application platforms are becoming increasingly popular,such as logistics delivery applications(SF Express,Meituan takeout)and real-time taxi applications(Didi Chuxing),etc.They all adopt spatial crowdsourcing technology to improve their service quality and play an increasingly important role in the daily life of the public.Task assignment algorithm is the key core technology in spatial crowdsourcing,which refers to the process of reasonably assigning crowdsourcing tasks to task executors,such as the assignment of taxi orders in online taxi platforms.Due to the characteristics of spatial crowdsourcing such as spatiotemporal mobility and dynamism,the task assignment algorithm of spatial crowdsourcing is more complex.Compared with traditional web-based crowdsourcing,spatiotemporal tasks and task executors are constrained by various offline spatiotemporal attributes,and most of the tasks and executors are dynamically changed in real time in practical applications,which make the spatiotemporal crowdsourcing task assignment problem more challenging.Considering the dynamic and real-time characteristics presented by spatial crowdsourcing task assignment,we abstract the spatial crowdsourcing task assignment problem into an online task assignment model.Different assignment models based on reinforcement learning methods are proposed for single task(single performer demand)and group task(multi-performer demand)with different amount of performer demands.For single task we propose an assignment method based on Markov Decision Process,which fully learns historical task assignment data,takes into account the long-term impact of current assignment decisions,and makes a better one-to-one matching decision of task and performer.For group task we propose a Multi-armed bandit based task assignment method.Facing unknown task performers learn online and observe the feedback results of performers on task assignment to optimize the assignment decision in real time.The degree of task acceptance by crowdsourced performers in a specific spatiotemporal environment is dynamically evaluated,resulting in a high acceptance rate when assigning tasks to a set of performers.The thesis uses the publicly available decision intelligence simulation platform of Didi and real Didi car dispatching dataset to conduct experiments on single task online assignment.The simulation results show that the single task assignment model proposed in this paper has a better performance in the scheduling of online taxi orders.For the task assignment problem in group task scenarios,simulation experiments are conducted based on the real Gowalla crowdsourcing check-in dataset,and the algorithm proposed in this paper is close to the optimal task assignment strategy in terms of assignment effect.We also develop a practical C2C space-time crowdsourcing system platform that supports mobile micro-tasking and adapts to the realistic needs of space-time crowdsourcing task assignment. |
