Long-Term or Temporary? Hybrid Worker Recruitment for Mobile Crowd Sensing and Computing

TL;DR

The paper addresses hybrid recruitment for mobile crowd sensing and computing under participation and workload uncertainties by combining offline long-term contracts with online temporary workers. It formulates the problem as a $0$-$1$ ILP with probabilistic constraints and proposes three algorithms—ESA, UISRFC, and GP-SCA—to obtain optimal or sub-optimal solutions, complemented by an online mode for temporary workers. The framework enables overbooking to hedge against uncertainties while managing risk and budgets, with extensive numerical and real-world dataset evaluations showing improved service quality and substantial time efficiency over baselines. The work advances practical MCSC provisioning by integrating offline planning with online adaptability and offers scalable algorithms with provable performance characteristics for NP-hard planning problems.

Abstract

This paper investigates a novel hybrid worker recruitment problem where the mobile crowd sensing and computing (MCSC) platform employs workers to serve MCSC tasks with diverse quality requirements and budget constraints, under uncertainties in workers' participation and their local workloads.We propose a hybrid worker recruitment framework consisting of offline and online trading modes. The former enables the platform to overbook long-term workers (services) to cope with dynamic service supply via signing contracts in advance, which is formulated as 0-1 integer linear programming (ILP) with probabilistic constraints of service quality and budget.Besides, motivated by the existing uncertainties which may render long-term workers fail to meet the service quality requirement of each task, we augment our methodology with an online temporary worker recruitment scheme as a backup Plan B to support seamless service provisioning for MCSC tasks, which also represents a 0-1 ILP problem. To tackle these problems which are proved to be NP-hard, we develop three algorithms, namely, i) exhaustive searching, ii) unique index-based stochastic searching with risk-aware filter constraint, iii) geometric programming-based successive convex algorithm, which achieve the optimal or sub-optimal solutions. Experimental results demonstrate our effectiveness in terms of service quality, time efficiency, etc.

Figures (8)

• Figure 1: Framework and timeline associated with the proposed hybrid worker recruitment for MCSC in wireless networks. The purple and yellow shadows describe the region of PoIs, while the dark-colored contract indicates the hard service quality assurance and the light-colored one means the soft quality assurance. Besides, the star aside each worker denotes the relevant local workload, where the darker color indicates a heavier local workload that may require more resources (e.g., storage, computing resources, etc.).
• Figure 2: Example of long-term contracts among 4 workers and 2 sensing tasks (the third subscript 1, and 0 denote "Hard", and "Soft", respectively for notational simplicity in this figure).
• Figure 3: (a). Performance comparison on the expected service quality of MCSC tasks upon considering small, and problem sizes under $\text{Set \#1}$; (b)$\sim$(e) Long-term contract signing associated with 2 tasks and 5 workers in (a); (f)$\sim$(h) Long-term contract signing associated with 5 tasks and 15 workers in (a). Note that figures (b)$\sim$(h), the yellow, and purple-colored box indicates soft, and hard service assurance, respectively.
• Figure 4: Performance comparison on the average practical service quality (per trading) of MCSC tasks upon considering different problem sizes, where (a) and (b) rely on $\text{Set \#1}$, while (c) and (d) rely on $\text{Set \#2}$. Note that the average value of each figure is derived from 300 trading.
• Figure 5: Performance comparison on $d_k^{B}$ and the average practical expense of MCSC tasks for purchasing workers' services (per trading), upon considering 2 tasks, 6 workers, and different values of budget. Specifically, (a) and (b) rely on $\text{Set \#1}$ upon considering $d^{B}_k\in[13,15]$ and $d^{B}_k\in[18,20]$; while (c) and (d) rely on $\text{Set \#2}$ upon considering $d^{B}_k\in[13,15]$ and $d^{B}_k\in[18,20]$. Note that the average value of each figure is derived from 300 trading.