LIMP: Large Language Model Enhanced Intent-aware Mobility Prediction
Songwei Li, Jie Feng, Jiawei Chi, Xinyuan Hu, Xiaomeng Zhao, Fengli Xu
TL;DR
The paper tackles the problem of predicting human mobility when underlying intentions are latent and not directly observable. It introduces LIMP, a three-pronged framework that combines an Analyze-Abstract-Infer (A2I) workflow to elicit intents from large language models, a fine-tuning pipeline to distill reasoning from GPT-4o into a smaller open-source model (Llama3-8B-Instruct), and a transformer-based mobility predictor that embeds probability-weighted intents $e_I$ into the trajectory modeling process, using $e_I = \sum_i e_{I_i} P(I_i|t,u)$. Empirically, LIMP achieves substantial gains over baselines on two real-world datasets (Beijing and Shanghai), including a relative Acc@1 improvement of about 6.6–9.5% and a notable ~16.3% boost in intent inference accuracy on the A2I task; ablations confirm the importance of probabilistic intent weighting and anchor-based home/work identification. The approach demonstrates scalable, interpretable, and cost-effective integration of LLM-based commonsense reasoning into domain-specific mobility prediction, with strong potential for real-world deployment and extension to other spatiotemporal tasks. All results are validated with extensive experiments and analyses, underscoring the practical value of coupling LLM-driven intent inference with a lightweight, intent-aware predictor.
Abstract
Human mobility prediction is essential for applications like urban planning and transportation management, yet it remains challenging due to the complex, often implicit, intentions behind human behavior. Existing models predominantly focus on spatiotemporal patterns, paying less attention to the underlying intentions that govern movements. Recent advancements in large language models (LLMs) offer a promising alternative research angle for integrating commonsense reasoning into mobility prediction. However, it is a non-trivial problem because LLMs are not natively built for mobility intention inference, and they also face scalability issues and integration difficulties with spatiotemporal models. To address these challenges, we propose a novel LIMP (LLMs for Intent-ware Mobility Prediction) framework. Specifically, LIMP introduces an "Analyze-Abstract-Infer" (A2I) agentic workflow to unleash LLM's commonsense reasoning power for mobility intention inference. Besides, we design an efficient fine-tuning scheme to transfer reasoning power from commercial LLM to smaller-scale, open-source language model, ensuring LIMP's scalability to millions of mobility records. Moreover, we propose a transformer-based intention-aware mobility prediction model to effectively harness the intention inference ability of LLM. Evaluated on two real-world datasets, LIMP significantly outperforms baseline models, demonstrating improved accuracy in next-location prediction and effective intention inference. The interpretability of intention-aware mobility prediction highlights our LIMP framework's potential for real-world applications. Codes and data can be found in https://github.com/tsinghua-fib-lab/LIMP .