Correlation-Based Task Processing Plans in Crowdsensing Platforms

Bo Yin, Jun Lu, Xuetao Wei

Abstract: Crowdsensing has been recognized as an innovative data gathering paradigm. For efficient management of workers and data collection, the physical area is often divided into multiple subareas where multiple crowdsensing tasks are involved in single subareas and workers in a subarea complete tasks independently and simultaneously. Previous studies only focused on planning task execution of individual workers, neglecting subarea-level task processing. From the perspective of crowdsensing platforms, it is critical to plan the task processing among different subareas, in order to promote crowdsensing performance by learning experience of preceding tasks. In this paper, we tackle the task processing plan problem that defines the inter-subarea execution order. We first introduce the correlation concept that quantifies the potential experience gained from consecutive execution according to task similarity. We then formalize the maximum-correlation processing plan problem that targets at generating a set of trees with the maximum value of total correlation. We prove that the maximum-correlation processing plan problem is equivalent to the directed maximum spanning tree problem over a modified correlation graph, and propose an efficient algorithm. We also refine the maximum-correlation processing plan for the trade-off between correlation and parallelism. Extensive evaluation shows that our proposed approaches achieve the promising results.