Dynamic Worker-and-Task Assignment on Uncertain Spatial Crowdsourcing

Abstract

Online spatial crowdsourcing is inherently dynamic and uncertainty. Static methods might fail when addressing crowdsourcing problems of the online environments where the optimization parameters are unknown in advance; and crowd-workers can join or leave the environments in runtime. Thus, the goal of a spatial task assignment is to deal with dynamic and uncertainty features inherent in online crowdsourcing environments. Moreover, dynamic spatial task assignment is further complicated when there are large-scale tasks and crowd-workers in crowdsourcing environments. Therefore in this paper, we propose a dynamic optimization algorithm based on Spatial-aware Multi-Agent Q-Learning (SMAQL), which can keep higher effectiveness in face of uncertainty and large-scale problems in complex crowdsourcing scenario. The proposed approach combines spatial optimization with multi-agent Q-Iearning. S-MAQL algorithm decomposes the crowdsourcing optimization problem into many sub-problems by means of a grid-based spatial optimization approach. And each agent utilizes Q-Iearning algorithms to handle its own region optimization problems of the real-time crowdsourcing. Through extensive experiments, we demonstrate the efficiency of the proposed algorithms on the real data sets.