HeteroStamp: leveraging heterogeneous social interactions for mobility prediction-enhanced cost-aware spatiotemporal crowdsensing

Accurately predicting user mobility is crucial for effectively assigning spatiotemporal crowdsensing tasks to appropriate mobile users, thereby enhancing task completion rates. While prior studies have proposed various trajectory-based mobility prediction methods, they have not adequately addressed the heterogeneous relationships between completed spatiotemporal tasks and users’ social trajectories. In this article, we present a new spatiotemporal crowdsensing framework that incorporates Heterogeneous Social interactions into the joint design of task allocation and mobility prediction (HeteroStamp). Unlike prior studies, HeteroStamp jointly considers heterogeneous social mobility prediction and social interaction cost-aware task allocation. To achieve this, HeteroStamp incorporates a new social mobility model that employs heterogeneous social graph embedding to extract historical mobility features. Furthermore, a tailored deep learning approach is utilized to predict users’ trajectories. Leveraging the mobility prediction module, HeteroStamp incorporates a cost-aware task allocation module that maximizes task coverage while considering users’ social interaction costs and tasks’ spatiotemporal completion constraints. We show the NP-hardness of the task coverage maximization problem, and propose both a local greedy algorithm and a global heuristic algorithm to efficiently solve it. To evaluate HeteroStamp’s performance, we conduct experiments using two widely-used real-world datasets. The results reveal that HeteroStamp outperforms existing state-of-the-art baselines in terms of achieving more accurate task-user matching and enhanced task completion rates.