Towards Optimized Online Task Allocation in Cost-Sensitive Crowdsensing Applications

In crowdsensing applications, participants (crowd sensors) work collectively to report their measurements about the physical world. This paper focuses on the optimized online task allocation problem in cost-sensitive crowdsensing applications where the goal is to dynamically allocate the sensing tasks to participants to meet the requirement of the applications while minimizing the sensing costs. Recent progress has been made to tackle the task allocation problem in crowdsensing. However, two important challenges have not been well addressed: i) 'physical dynamics': the values of the measured variables in crowdsensing often change significantly over time and space. It is essential for the task allocation schemes to adapt to such changes efficiently to optimize the task allocation process; ii) 'crowd irregularity': the number of participants in crowdsensing is often smaller than the number of desirable sensing locations and not all crowd sensors contribute data all the time (e.g., due to incentive or budget constraints). To address the above challenges, this paper develops an Online Optimized Task Allocation (OO-TA) scheme inspired by techniques from information theory and online learning. We evaluate the OO-TA scheme using a dataset collected from a real-world crowdsensing application. The evaluation results show that OO-TA scheme significantly outperforms the state-of-the-art baselines in terms of both effectiveness and efficiency.