Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ITINERA: Integrating Spatial Optimization with Large Language Models for Open-domain Urban Itinerary Planning

Yihong Tang, Zhaokai Wang, Ao Qu, Yihao Yan, Zhaofeng Wu, Dingyi Zhuang, Jushi Kai, Kebing Hou, Xiaotong Guo, Han Zheng, Tiange Luo, Jinhua Zhao, Zhan Zhao, Wei Ma

TL;DR

This work formalizes Open-domain Urban Itinerary Planning (OUIP) and introduces ITINERA, a system that fuses large language models with spatial optimization to produce personalized, spatially coherent city itineraries from natural language requests. It decomposes requests, retrieves preferred POIs from a user-owned database, clusters and orders POIs via a hierarchical TSP approach, and generates final itineraries with an LLM under explicit constraints. Key contributions include a formal OUIP problem definition, a five-module pipeline (UPC, RD, PPR, CSO, IG), new evaluation metrics, and extensive real-world experiments plus a deployed system demonstrating superior personalization and spatial coherence over baselines. The results indicate significant improvements in user alignment and travel efficiency, highlighting the practical potential of integrating LLMs with spatial optimization for urban planning tasks.

Abstract

Citywalk, a recently popular form of urban travel, requires genuine personalization and understanding of fine-grained requests compared to traditional itinerary planning. In this paper, we introduce the novel task of Open-domain Urban Itinerary Planning (OUIP), which generates personalized urban itineraries from user requests in natural language. We then present ITINERA, an OUIP system that integrates spatial optimization with large language models to provide customized urban itineraries based on user needs. This involves decomposing user requests, selecting candidate points of interest (POIs), ordering the POIs based on cluster-aware spatial optimization, and generating the itinerary. Experiments on real-world datasets and the performance of the deployed system demonstrate our system's capacity to deliver personalized and spatially coherent itineraries compared to current solutions. Source codes of ITINERA are available at https://github.com/YihongT/ITINERA.

ITINERA: Integrating Spatial Optimization with Large Language Models for Open-domain Urban Itinerary Planning

TL;DR

This work formalizes Open-domain Urban Itinerary Planning (OUIP) and introduces ITINERA, a system that fuses large language models with spatial optimization to produce personalized, spatially coherent city itineraries from natural language requests. It decomposes requests, retrieves preferred POIs from a user-owned database, clusters and orders POIs via a hierarchical TSP approach, and generates final itineraries with an LLM under explicit constraints. Key contributions include a formal OUIP problem definition, a five-module pipeline (UPC, RD, PPR, CSO, IG), new evaluation metrics, and extensive real-world experiments plus a deployed system demonstrating superior personalization and spatial coherence over baselines. The results indicate significant improvements in user alignment and travel efficiency, highlighting the practical potential of integrating LLMs with spatial optimization for urban planning tasks.

Abstract

Citywalk, a recently popular form of urban travel, requires genuine personalization and understanding of fine-grained requests compared to traditional itinerary planning. In this paper, we introduce the novel task of Open-domain Urban Itinerary Planning (OUIP), which generates personalized urban itineraries from user requests in natural language. We then present ITINERA, an OUIP system that integrates spatial optimization with large language models to provide customized urban itineraries based on user needs. This involves decomposing user requests, selecting candidate points of interest (POIs), ordering the POIs based on cluster-aware spatial optimization, and generating the itinerary. Experiments on real-world datasets and the performance of the deployed system demonstrate our system's capacity to deliver personalized and spatially coherent itineraries compared to current solutions. Source codes of ITINERA are available at https://github.com/YihongT/ITINERA.
Paper Structure (39 sections, 3 equations, 4 figures, 5 tables, 4 algorithms)

This paper contains 39 sections, 3 equations, 4 figures, 5 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. LLMs in Urban Applications
  4. Itinerary Planning (IP)
  5. Methodology
  6. Open-domain Urban Itinerary Planning (OUIP) Problem
  7. User-owned POI Database Construction
  8. Request Decomposition
  9. Preference-aware POI Retrieval
  10. Cluster-aware Spatial Optimization
  11. Spatial Clustering and Filtering
  12. POI Ordering via Solving Hierarchical TSP
  13. Itinerary Generation
  14. Experiments
  15. Experimental Setup
  16. ...and 24 more sections

Figures (4)

  • Figure 1: The OUIP problem and the OUIP system.
  • Figure 2: An overview of the proposed ItiNera system.
  • Figure 3: Generated itineraries of ItiNera (left) and GPT-4 CoT (right).
  • Figure 4: Screenshots of the deployed system: POI view & Itinerary view.