Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MAS4POI: a Multi-Agents Collaboration System for Next POI Recommendation

Yuqian Wu, Yuhong Peng, Jiapeng Yu, Raymond S. T. Lee

TL;DR

MAS4POI addresses the complexity of next POI recommendation by deploying seven specialized LLM-based agents that collaborate to preprocess data, propose initial recommendations, refine them, answer queries, and plan navigation. The approach integrates DataAgent, Manager, Analyst, Reflector, UserAgent, Searcher, and Navigator into a three-application workflow (Next POI, Q&A, Navigation) and is evaluated on NYC and TKY datasets with six LLMs, showing improved accuracy and reduced cold-start impact. The key contributions include a flexible MAS4POI framework, iterative refinement via Reflector, and empirical validation across real-world data, with public code release. The work has practical significance for real-time location-based services and multimodal decision-making using LLMs.

Abstract

LLM-based Multi-Agent Systems have potential benefits of complex decision-making tasks management across various domains but their applications in the next Point-of-Interest (POI) recommendation remain underexplored. This paper proposes a novel MAS4POI system designed to enhance next POI recommendations through multi-agent interactions. MAS4POI supports Large Language Models (LLMs) specializing in distinct agents such as DataAgent, Manager, Analyst, and Navigator with each contributes to a collaborative process of generating the next POI recommendations.The system is examined by integrating six distinct LLMs and evaluated by two real-world datasets for recommendation accuracy improvement in real-world scenarios. Our code is available at https://github.com/yuqian2003/MAS4POI.

MAS4POI: a Multi-Agents Collaboration System for Next POI Recommendation

TL;DR

MAS4POI addresses the complexity of next POI recommendation by deploying seven specialized LLM-based agents that collaborate to preprocess data, propose initial recommendations, refine them, answer queries, and plan navigation. The approach integrates DataAgent, Manager, Analyst, Reflector, UserAgent, Searcher, and Navigator into a three-application workflow (Next POI, Q&A, Navigation) and is evaluated on NYC and TKY datasets with six LLMs, showing improved accuracy and reduced cold-start impact. The key contributions include a flexible MAS4POI framework, iterative refinement via Reflector, and empirical validation across real-world data, with public code release. The work has practical significance for real-time location-based services and multimodal decision-making using LLMs.

Abstract

LLM-based Multi-Agent Systems have potential benefits of complex decision-making tasks management across various domains but their applications in the next Point-of-Interest (POI) recommendation remain underexplored. This paper proposes a novel MAS4POI system designed to enhance next POI recommendations through multi-agent interactions. MAS4POI supports Large Language Models (LLMs) specializing in distinct agents such as DataAgent, Manager, Analyst, and Navigator with each contributes to a collaborative process of generating the next POI recommendations.The system is examined by integrating six distinct LLMs and evaluated by two real-world datasets for recommendation accuracy improvement in real-world scenarios. Our code is available at https://github.com/yuqian2003/MAS4POI.
Paper Structure (21 sections, 15 equations, 4 figures, 4 tables, 1 algorithm)

This paper contains 21 sections, 15 equations, 4 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Statement
  4. Method
  5. Framework Overview.
  6. Role Agents.
  7. Manager
  8. Reflector
  9. DataAgent
  10. Navigator
  11. Analyst
  12. UserAgent
  13. Searcher
  14. Experiments
  15. Experiment Setup
  16. ...and 6 more sections

Figures (4)

  • Figure 1: Illustration of a user's historical trajectory (solid line) and the candidate POIs for the next visit (dashed line).
  • Figure 2: The Overall Framework of MAS4POI in Next POI Recommendation Task.
  • Figure 3: MAS4POI Workflow with Key Elements Highlighted (Red: incorrect POI recommendations initially generated by Analyst, Blue: Refined POI recommendation output, Orange: REFLECTION process carried out by Reflector, Green: API Results, Purple: User Requests)
  • Figure 4: Upper: This table shows the improvement in $Acc@k$ and MRR with different states based on GPT-3.5-Turbo, when state equals to $y_{0}$, it indicates no use of Reflector. Below: This shows performance improvements with iterations.