User selection utilizing data properties in mobile crowdsensing

Abstract

The information sensed by a mobile device is just data, while the aggregation of information sensed by thousands of devices is knowledge. This is the basic idea of Mobile CowdSensing (MCS). In MCS, the user selection problem has always been a key issue, where the task publisher attempts to recruit a suitable set of users to cooperatively sense the events or data at some particular time points and places. Hence, the temporal and spatial constraint becomes the main criterion to select the users. However, whether a user prefers to accomplish the sensing data not only depends on the temporal and spatial relationship, but also lies on the data property. In other words, a user is likely to take the sensing task if it is interested in the data to be sensed. In this paper, we propose a user Selection utilizing data properties in mobile crowdsensing (SPM), where a triple-layer structure considering not only the temporal and spatial probability, but also the data’s property is formulated, and task finishing probabilities are calculated in both the intentional and unintentional situations. Then, a greedy algorithm is proposed to select a suitable set of users for finishing the sensing tasks. We conduct extensive simulations based on three widely-used real-world traces: geolife, roma/taxi, epfl and a synthetic data set. The results show that, compared with other user selection strategies, SPM finishes the largest number of sensing tasks.