Reinforced Reliable Worker Selection for Spatial Crowdsensing Networks

Abstract

Spatial Crowdsensing Networks limit the sensing tasks in some special places where workers should sense data for them. Due to the lack of a priori information about quality of workers, guaranteeing the quality of the sensing tasks remains a key challenge. In this paper, we model the quality of workers through two factors, namely bias and variance, which describe the continuous value feature of sensing tasks. After calibrating the bias, we should iteratively estimate worker variances more and more accurately. Meanwhile, we should select more reliable workers with low variances to finish sensing tasks. This is a classic exploration and exploitation dilemma. Therefore, to overcome the dilemma, we design a novel Multi-Armed Bandit (MAB) algorithm which is based on Upper Confidence Bounds (UCB) scheme and combined with a weighted data aggregation scheme to calculate a better ground truth of a sensing task. Then, we prove the expected sensing error of sensing tasks can be bounded according to the regret bound of the MAB in our setting. In simulation experiments, we use a real world data set to validate the theoretical results of our algorithm and it outperforms two baselines significantly in different settings.