Causality-Aware Next Location Prediction Framework based on Human Mobility Stratification

TL;DR

This work tackles next location prediction by addressing causal confounding in human mobility data through a causality-aware framework that stratifies travels into anchor-targeted and nonanchor-targeted patterns. It introduces a DAG-based causal formulation with direct ($H\to Y$, $L\to Y$) and indirect ($H\to G\to Y$) pathways and employs counterfactuals $do(G=g^*)$ to separate effects, enabling a plug-and-play predictor that uses two hidden states $h$ and $g$ and an MLP to produce $\hat{y}_{pred}$ and $\hat{y}_{causal}$ for nonanchor travels. The model is trained with a multi-task objective across anchor- and nonanchor-targeted trajectories and evaluated on Foursquare (Tokyo, New York) and Blogwatcher datasets, showing notable gains in Recall@5, MRR@5, and NDCG@5 over strong baselines; ablations confirm the necessity of the direct causal links and the utility of stratification. Overall, the approach improves prediction accuracy and interpretability in location-based services by explicitly modeling causal structure and indirect effects in human mobility.

Abstract

Human mobility data are fused with multiple travel patterns and hidden spatiotemporal patterns are extracted by integrating user, location, and time information to improve next location prediction accuracy. In existing next location prediction methods, different causal relationships that result from patterns in human mobility data are ignored, which leads to confounding information that can have a negative effect on predictions. Therefore, this study introduces a causality-aware framework for next location prediction, focusing on human mobility stratification for travel patterns. In our research, a novel causal graph is developed that describes the relationships between various input variables. We use counterfactuals to enhance the indirect effects in our causal graph for specific travel patterns: non-anchor targeted travels. The proposed framework is designed as a plug-and-play module that integrates multiple next location prediction paradigms. We tested our proposed framework using several state-of-the-art models and human mobility datasets, and the results reveal that the proposed module improves the prediction performance. In addition, we provide results from the ablation study and quantitative study to demonstrate the soundness of our causal graph and its ability to further enhance the interpretability of the current next location prediction models.

Figures (7)

• Figure 1: Two travel patterns.
• Figure 2: Statistical impacts of knowing previous location or not on the next location predictions.
• Figure 3: Causal graph of next location prediction: (a) conventional, and (b) our approach.. Observations $\{U$, $L$, $T\}$ in (a) and (b) refer to user, location and time information; $\{H, G\}$: hidden states; $Y$: next predicted location
• Figure 4: Calculating causal effects through counterfactual
• Figure 5: Structure of causality-aware next location prediction framework. Counterfactuals will replace the original hidden state $\textbf{g}$. Then, intervened prediction results with $\textbf{g*}$ will be used for training by calculating $\hat{y}_{causal}$ but will not exist during validation and test processes.

Theorems & Definitions (3)

• Definition 1: Human Mobility Data
• Definition 2: Anchor Locations
• Definition 3: Human Mobility Stratification