Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CausalMob: Causal Human Mobility Prediction with LLMs-derived Human Intentions toward Public Events

Xiaojie Yang, Hangli Ge, Jiawei Wang, Zipei Fan, Renhe Jiang, Ryosuke Shibasaki, Noboru Koshizuka

TL;DR

CausalMob addresses the challenge of predicting human mobility under irregular public events by introducing a causal framework that leverages LLM-derived human intentions extracted from news articles. It combines multidimensional event treatments with confounder learning via ST-ResNET-based representations and a GRU-based predictor that incorporates a reweighting mechanism and an IPM-bounded counterfactual loss. The model demonstrates state-of-the-art predictive performance on Japanese regional mobility while providing interpretable analyses of event effects and potential policy implications. This approach offers a principled way to quantify and interpret how public events influence mobility, enabling proactive decision-making in urban management and disaster response.

Abstract

Large-scale human mobility exhibits spatial and temporal patterns that can assist policymakers in decision making. Although traditional prediction models attempt to capture these patterns, they often interfered by non-periodic public events, such as disasters and occasional celebrations. Since regular human mobility patterns are heavily affected by these events, estimating their causal effects is critical to accurate mobility predictions. Although news articles provide unique perspectives on these events in an unstructured format, processing is a challenge. In this study, we propose a causality-augmented prediction model, called CausalMob, to analyze the causal effects of public events. We first utilize large language models (LLMs) to extract human intentions from news articles and transform them into features that act as causal treatments. Next, the model learns representations of spatio-temporal regional covariates from multiple data sources to serve as confounders for causal inference. Finally, we present a causal effect estimation framework to ensure event features remain independent of confounders during prediction. Based on large-scale real-world data, the experimental results show that the proposed model excels in human mobility prediction, outperforming state-of-the-art models.

CausalMob: Causal Human Mobility Prediction with LLMs-derived Human Intentions toward Public Events

TL;DR

CausalMob addresses the challenge of predicting human mobility under irregular public events by introducing a causal framework that leverages LLM-derived human intentions extracted from news articles. It combines multidimensional event treatments with confounder learning via ST-ResNET-based representations and a GRU-based predictor that incorporates a reweighting mechanism and an IPM-bounded counterfactual loss. The model demonstrates state-of-the-art predictive performance on Japanese regional mobility while providing interpretable analyses of event effects and potential policy implications. This approach offers a principled way to quantify and interpret how public events influence mobility, enabling proactive decision-making in urban management and disaster response.

Abstract

Large-scale human mobility exhibits spatial and temporal patterns that can assist policymakers in decision making. Although traditional prediction models attempt to capture these patterns, they often interfered by non-periodic public events, such as disasters and occasional celebrations. Since regular human mobility patterns are heavily affected by these events, estimating their causal effects is critical to accurate mobility predictions. Although news articles provide unique perspectives on these events in an unstructured format, processing is a challenge. In this study, we propose a causality-augmented prediction model, called CausalMob, to analyze the causal effects of public events. We first utilize large language models (LLMs) to extract human intentions from news articles and transform them into features that act as causal treatments. Next, the model learns representations of spatio-temporal regional covariates from multiple data sources to serve as confounders for causal inference. Finally, we present a causal effect estimation framework to ensure event features remain independent of confounders during prediction. Based on large-scale real-world data, the experimental results show that the proposed model excels in human mobility prediction, outperforming state-of-the-art models.

Paper Structure

This paper contains 25 sections, 10 equations, 12 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Data and Analysis
  4. Observational Data
  5. Preliminary Data Analysis
  6. Problem Definition
  7. Methodology
  8. Human Intentions Extraction with Semantic-Aware LLMs
  9. Causal Effect Inference through Counterfactual Estimation
  10. Confounder Representation Learning by Spatio-Temporal Data Modeling
  11. Human Mobility Prediction Enhanced by Causality Modeling
  12. Experiment
  13. Experimental Settings
  14. Overall Performance
  15. Analysis of LLMs-derived Human Intentions
  16. ...and 10 more sections

Figures (12)

  • Figure 1: Conceptual graph of this research
  • Figure 2: Causal graph of this research
  • Figure 3: Public events examples along with changes of daily human mobility. We calculate mean values and standard deviations of daily human mobility records ($\star$ markers refer to the dates when public events happened).
  • Figure 4: Number of News labeled with categories
  • Figure 5: Framework of CausalMob: We first process news articles to obtain structured information of public events in part (a). Then in part (b) we use ST-ResNET to learn the representation of the confounder $z$. Finally, an inference network and a re-weighting network are built to calculate factual loss and counterfactual loss as causal inference.
  • ...and 7 more figures