LC-TDC: A low cost and truth data collection scheme by using missing data imputation in sparse mobile crowdsensing

Abstract

Mobile Crowd Sensing (MCS) is a promising computing paradigm for data collection harnessing ubiquitous workers equipped with sensing devices. While some studies have delved into quality-based and truthful data collection in sparse MCS, a research gap persists in the domain of low-cost and truthful data collection for the Post-Unknown Worker Recruitment (PUWR) problem, where the sensing qualities of workers remain unknown even after the platform acquires their data. We first attempt to tackle these challenges by proposing a Low Cost and Truth Data Collection (LC-TDC) scheme. Firstly, the Deep Matrix Factorization (DMF) model is utilized to infer data, facilitating the evaluation of workers' trustworthiness and missing data imputation. This approach not only prevents the acceptance of erroneous or malicious data but also contributes to the reduction of data collection costs. Secondly, we propose a method to recalibrate worker quality based on historical data, accelerating worker identification at almost no cost. Thirdly, a profit optimization algorithm is proposed to determine the optimal number of verified workers according to the platform's accuracy in identifying workers, which can maximize the platform's utility in the long term. Extensive experimental results demonstrate that the proposed LC-TDC scheme outperforms previous strategies in terms of data accuracy, cost, and platform profit.