The rapid growth of mobile networks and widespread mobile device usage has brought mobile crowdsourcing (MCS) for location-sensitive services into focus, particularly for spatial task assignments. However, current research often overlooks the combined impact of users’ category and mobile preferences, resulting in suboptimal task assignments. Additionally, the inherent heterogeneity among users is frequently ignored, failing to represent the true dynamics of the MCS ecosystem. To address these gaps, we propose a comprehensive framework, Task Assignment with User Preference and Heterogeneity, consisting of two key components: the Category and Mobile Combined Preference (CAMP) model and the Preference-Aware Task Assignment mechanism. The CAMP model predicts users’ combined category and mobile preferences using attention mechanisms to extract insights from sparse historical data. In parallel, the Preference-Aware Task Assignment mechanism introduces three novel algorithms that account for user preference and capacity heterogeneity. Extensive experiments on real-world datasets demonstrate the effectiveness and efficiency of the proposed methods.