UniMove: A Unified Model for Multi-city Human Mobility Prediction

TL;DR

UniMove tackles cross-city human mobility prediction by introducing a trajectory-location dual-tower model that shares parameters across cities while learning universal spatial representations. The Location Tower encodes grid-based spatial features (POI distribution, coordinates, popularity) and the Trajectory Tower uses MoE Transformer blocks to model heterogeneous city-specific mobility patterns. Joint training on multi-city data yields mutual enhancement and significant accuracy gains over state-of-the-art baselines, demonstrating the feasibility of a unified foundational model for mobility. The work provides thorough ablation, scalability, and MoE analyses and releases the implementation for public use.

Abstract

Human mobility prediction is vital for urban planning, transportation optimization, and personalized services. However, the inherent randomness, non-uniform time intervals, and complex patterns of human mobility, compounded by the heterogeneity introduced by varying city structures, infrastructure, and population densities, present significant challenges in modeling. Existing solutions often require training separate models for each city due to distinct spatial representations and geographic coverage. In this paper, we propose UniMove, a unified model for multi-city human mobility prediction, addressing two challenges: (1) constructing universal spatial representations for effective token sharing across cities, and (2) modeling heterogeneous mobility patterns from varying city characteristics. We propose a trajectory-location dual-tower architecture, with a location tower for universal spatial encoding and a trajectory tower for sequential mobility modeling. We also design MoE Transformer blocks to adaptively select experts to handle diverse movement patterns. Extensive experiments across multiple datasets from diverse cities demonstrate that UniMove truly embodies the essence of a unified model. By enabling joint training on multi-city data with mutual data enhancement, it significantly improves mobility prediction accuracy by over 10.2\%. UniMove represents a key advancement toward realizing a true foundational model with a unified architecture for human mobility. We release the implementation at https://github.com/tsinghua-fib-lab/UniMove/.

Figures (6)

• Figure 1: Comparison between existing solutions and our approach.
• Figure 2: The overall architecture of UniMove. (a)Trajectory Tower and Location Tower:Trajectory Tower predicts the latent embedding of the next location based on historical trajectories, Location Tower generates the feature representations for all candidate locations. (b)MoE(Mixture of Experts) Transformer Block. (c)Location Encoder. (d)Deep & Cross Net for further capturing features from different locations.
• Figure 3: Locations embedding of three cities based on t-SNE visualization. (a)-(c) are original embeddings, and (d)-(f) are final embeddings learned by DCN.
• Figure 4: Expert selection distribution in MOE Layers across three city datasets.
• Figure 5: Data scalability(left) and model scalability(right).