Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Semantic Trajectory Data Mining with LLM-Informed POI Classification

Yifan Liu, Chenchen Kuai, Haoxuan Ma, Xishun Liao, Brian Yueshuai He, Jiaqi Ma

TL;DR

This paper tackles the lack of semantic context in GPS-based trajectory data by introducing a two-stage framework that first uses large language models to classify POIs and then applies a Bayesian approach to infer stay-point activities. The method is designed to work with incomplete POI features and without additional training, demonstrating robustness across diverse datasets (LA County and Egypt) using OpenStreetMap data. Key contributions include the first application of LLMs to POI classification, a probabilistic activity inference mechanism that combines nearby POIs and temporal priors, and strong empirical results (POI classification accuracy up to 93.4% and activity inference accuracy around 91–92%). The work advances semantic trajectory mining with practical implications for transportation systems, local searches, and mobility research, while addressing data quality challenges and privacy considerations.

Abstract

Human travel trajectory mining is crucial for transportation systems, enhancing route optimization, traffic management, and the study of human travel patterns. Previous rule-based approaches without the integration of semantic information show a limitation in both efficiency and accuracy. Semantic information, such as activity types inferred from Points of Interest (POI) data, can significantly enhance the quality of trajectory mining. However, integrating these insights is challenging, as many POIs have incomplete feature information, and current learning-based POI algorithms require the integrity of datasets to do the classification. In this paper, we introduce a novel pipeline for human travel trajectory mining. Our approach first leverages the strong inferential and comprehension capabilities of large language models (LLMs) to annotate POI with activity types and then uses a Bayesian-based algorithm to infer activity for each stay point in a trajectory. In our evaluation using the OpenStreetMap (OSM) POI dataset, our approach achieves a 93.4% accuracy and a 96.1% F-1 score in POI classification, and a 91.7% accuracy with a 92.3% F-1 score in activity inference.

Semantic Trajectory Data Mining with LLM-Informed POI Classification

TL;DR

This paper tackles the lack of semantic context in GPS-based trajectory data by introducing a two-stage framework that first uses large language models to classify POIs and then applies a Bayesian approach to infer stay-point activities. The method is designed to work with incomplete POI features and without additional training, demonstrating robustness across diverse datasets (LA County and Egypt) using OpenStreetMap data. Key contributions include the first application of LLMs to POI classification, a probabilistic activity inference mechanism that combines nearby POIs and temporal priors, and strong empirical results (POI classification accuracy up to 93.4% and activity inference accuracy around 91–92%). The work advances semantic trajectory mining with practical implications for transportation systems, local searches, and mobility research, while addressing data quality challenges and privacy considerations.

Abstract

Human travel trajectory mining is crucial for transportation systems, enhancing route optimization, traffic management, and the study of human travel patterns. Previous rule-based approaches without the integration of semantic information show a limitation in both efficiency and accuracy. Semantic information, such as activity types inferred from Points of Interest (POI) data, can significantly enhance the quality of trajectory mining. However, integrating these insights is challenging, as many POIs have incomplete feature information, and current learning-based POI algorithms require the integrity of datasets to do the classification. In this paper, we introduce a novel pipeline for human travel trajectory mining. Our approach first leverages the strong inferential and comprehension capabilities of large language models (LLMs) to annotate POI with activity types and then uses a Bayesian-based algorithm to infer activity for each stay point in a trajectory. In our evaluation using the OpenStreetMap (OSM) POI dataset, our approach achieves a 93.4% accuracy and a 96.1% F-1 score in POI classification, and a 91.7% accuracy with a 92.3% F-1 score in activity inference.
Paper Structure (18 sections, 4 equations, 5 figures, 5 tables)

This paper contains 18 sections, 4 equations, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. POI Classification
  4. Activity Inference
  5. Methodology
  6. Problem Formulation
  7. POI Classification
  8. Activity Inference
  9. Experiment
  10. Dataset
  11. Evaluation Setting
  12. POI classification
  13. Activity Inference
  14. Result
  15. POI Classification
  16. ...and 3 more sections

Figures (5)

  • Figure 1: Architecture of the proposed framework from POI classification (left) to activity Inference (right).
  • Figure 2: Example of the prompt and output from LLM.
  • Figure 3: Flow chart of activity inference procedure.
  • Figure 4: Example of POIs and classification results from (a) LA county and (b) Egypt.
  • Figure 5: Examples of semantic trajectories post-data mining: (a) Activity-annotated stay points in LA. (b) Activity timelines.