Near-Optimal User Recruitment in Mobile Crowdsensing for Urban Fine-Grained Event Detection

Thanks to the popularization of mobile smart devices equipped with various sensors like smartphones, the concept of mobile crowdsensing has come forth as a promising data collecting paradigm. Event detection in urban areas (i.e., traffic jam monitoring) is an important application of mobile crowdsensing, which can be implemented by recruiting a set of smart device users to collect plenty of fine-grained sensing data. However, as users are mobile and their sensing data are unreliable, it is hard to ensure that all events can be detected accurately. Thus, which users are recruited should be carefully determined to achieve a high detection accuracy and control the costs of users within a given budget. Unfortunately, we prove that the user recruitment problem in mobile crowdsensing for event detection is a NP-hard problem, indicating that there is no polynomial-time algorithm to achieve the optimal solution unless P = NP. In this work, we propose a polynomial-time near-optimal user recruitment algorithm, by leveraging the properties of adaptive monotonicity and adaptive submodularity. Our algorithm is theoretically proved to achieve a constant approximation ratio, compared with the optimum. Moreover, a data-dependent upper bound of our solution is also derived, providing a tighter performance guarantee. We also provide an accelerated version of our proposed algorithm by reducing its computation load. Extensive simulations are conducted, which show our proposed algorithm outperforms baselines under different settings and achieves near-optimal performance. Besides, the execution time of the accelerated version is significantly reduced.