Fused User Preference Learning for Task Assignment in Mobile Crowdsourcing

Abstract

With the development of GPS-enabled smart devices and wireless networks, mobile crowdsourcing (MCS) has received wide attention in assigning location-sensitive tasks to mobile users. The task assignment problem, in which tasks are released on the platforms and then assigned to available users, is a fundamental problem in MCS. However, existing works generally consider users’ category preference and mobile preference separately. Ignorance of the correlation between them could lead to poor assignment results. To this end, We propose a framework, Task Assignment with User Preference Learning, which consists of two components: 1) Fused User Preference Learning (FUP); and 2) Preference-Based Task Assignment. The first component called FUP is a fusion of task-category preference learning and spatial-temporal preference learning. For task-category preference learning, we propose a graph session-based learning model with attention components to exploit users’ sparse historical records. To our knowledge, we are the first to use a graph session-based learning model to explore task-category preference in MCS. Meanwhile, we propose an efficient function metric to characterize the spatial-temporal preference of users. The second component aims to achieve effective task assignment, in which we give higher priorities to users with higher preference scores for the tasks. Extensive evaluations of real data show the effectiveness and efficiency of the proposed solutions.