AdaTaskRec: An Adaptive Task Recommendation Framework in Spatial Crowdsourcing

Spatial crowdsourcing is one of the prime movers for the orchestration of location-based tasks, and task recommendation is a crucial means to help workers discover attractive tasks. While a number of existing studies have focused on modeling workers' geographical preferences in task recommendation, they ignore the phenomenon of workers' travel intention drifts across geographical areas, i.e., workers tend to have different intentions when they travel in different areas, which discounts the task recommendation quality of existing methods especially for workers that travel in unfamiliar out-of-town areas. To address this problem, we propose an Adaptive Task Recommendation (AdaTaskRec) framework. Specifically, we first give a novel two-module worker preference learning architecture that can calculate workers' preferences for POIs (that tasks are associated with) in different areas adaptively based on workers' current locations. If we detect that a worker is in the hometown area, then we apply the hometown preference learning module, which hybrids different strategies to aggregate workers' travel intentions into their preferences while considering the transition and the sequence patterns among locations. Otherwise, we invoke the out-of-town preference learning module, which is to capture workers' preferences by learning their travel intentions and transferring their hometown preferences into their out-of-town ones. Additionally, to improve task recommendation effectiveness, we propose a dynamic top-k recommendation method that sets different k values dynamically according to the numbers of neighboring workers and tasks. We also give an extra-reward-based and a fair top-k recommendation method, which introduce the extra rewards for tasks based on their recommendation rounds and consider exposure-based fairness of tasks, respectively. Extensive experiments offer insight into the effectiveness of the proposed framework.