GeoGen: A Two-stage Coarse-to-Fine Framework for Fine-grained Synthetic Location-based Social Network Trajectory Generation

TL;DR

This work tackles the challenge of producing high-fidelity, fine-grained synthetic LBSN trajectories under privacy constraints. It introduces GeoGen, a two-stage coarse-to-fine framework that first reconstructs spatially continuous, temporally regular latent movement sequences with a Sparsity-aware Spatio-temporal Diffusion model (S$^2$TDiff) and its SASG-UNet denoiser, then generates fine-grained trajectories using Coarse2FineNet, a Transformer-based Seq2Seq model with a Dynamic Context Fusion encoder and a Multi-task Hybrid-Head decoder that includes a Neural Temporal Point Process for timestamps. The approach yields superior fidelity (measured by JSDs across Distance, Radius, Interval, Length) and utility (next check-in prediction accuracy) across four real-world datasets, outperforming eight baselines and demonstrating robust generalization. Key technical contributions include the S$^2$TDiff diffusion framework with a sparsity-aware denoiser, the S$^2$G Attention mechanism, the POI-contextualized dynamic encoder, and the cross-task decoder that jointly models POI choice and non-uniform timestamps. The framework enables scalable, privacy-preserving generation of large-scale LBSN trajectory data with practical downstream utility for tasks like next-location prediction and mobility analysis.

Abstract

Location-Based Social Network (LBSN) check-in trajectory data are important for many practical applications, like POI recommendation, advertising, and pandemic intervention. However, the high collection costs and ever-increasing privacy concerns prevent us from accessing large-scale LBSN trajectory data. The recent advances in synthetic data generation provide us with a new opportunity to achieve this, which utilizes generative AI to generate synthetic data that preserves the characteristics of real data while ensuring privacy protection. However, generating synthetic LBSN check-in trajectories remains challenging due to their spatially discrete, temporally irregular nature and the complex spatio-temporal patterns caused by sparse activities and uncertain human mobility. To address this challenge, we propose GeoGen, a two-stage coarse-to-fine framework for large-scale LBSN check-in trajectory generation. In the first stage, we reconstruct spatially continuous, temporally regular latent movement sequences from the original LBSN check-in trajectories and then design a Sparsity-aware Spatio-temporal Diffusion model (S$^2$TDiff) with an efficient denosing network to learn their underlying behavioral patterns. In the second stage, we design Coarse2FineNet, a Transformer-based Seq2Seq architecture equipped with a dynamic context fusion mechanism in the encoder and a multi-task hybrid-head decoder, which generates fine-grained LBSN trajectories based on coarse-grained latent movement sequences by modeling semantic relevance and behavioral uncertainty. Extensive experiments on four real-world datasets show that GeoGen excels state-of-the-art models for both fidelity and utility evaluation, e.g., it increases over 69% and 55% in distance and radius metrics on the FS-TKY dataset.

Figures (7)

• Figure 1: Example of a discrete LBSN trajectory with irregular intervals (blue); and a continuous GPS trajectory with a fixed interval of 30 seconds (red).
• Figure 2: The overall architecture of our proposed GeoGen framework consists of two main stages. In the first stage, within S$^2$TDiff, each reconstructed latent movement sequence $s_{\text{lat}}$ is first transformed into pure noise $s_{\text{lat}}^N$. During inference, SASG-UNet progressively denoises a sampled noise sequence to generate a synthetic latent trajectory. In the second stage, the context-aware encoder of Coarse2FineNet extracts contextualized representations from the generated latent movement sequence. The decoder then auto-regressively generates each POI and fine-grained timestamp pair, conditioned on both the contextual representations and the partially generated LBSN check-in trajectory. FFN and TPP refer to the feed-forward network and temporal point process, respectively.
• Figure 3: The architecture of the proposed Spatially-Aware Sparsely-Gated U-Net (SASG-UNet). The kernel sizes for the average pooling layers in the Hierarchical 1D Convolution layers are indicated in parentheses.
• Figure 4: Performance comparison of different methods on FS-NYC dataset.
• Figure 5: CDF curves of interpolated LBSN check-in trajectories with different intervals.