PEMANT: Persona-Enriched Multi-Agent Negotiation for Travel
Summary: arXiv:2604.10475v1 Announce Type: new.
Abstract
Modeling household-level trip generation is fundamental to accurate demand forecasting, traffic flow estimation, and urban system planning. Existing studies have primarily relied on classical machine learning models, which exhibit limited predictive capability. Recent approaches utilizing large language models (LLMs) have not yet integrated behavioral theory or intra-household interaction dynamics, both essential for modeling realistic collective travel decisions.
Introduction to PEMANT
To address these limitations, we introduce a novel LLM-based framework named Persona-Enriched Multi-Agent Negotiation for Travel (PEMANT). This innovative framework first incorporates behavioral theory for individualized persona modeling and subsequently performs household-level trip planning negotiations through a structured multi-agent conversation.
Key Features of PEMANT
- Behavioral Theory Integration: PEMANT transforms static sociodemographic attributes into coherent narrative profiles. These profiles explicitly encode household-level attitudes, subjective norms, and perceived behavioral controls.
- Household-Aware Chain-of-Planned-Behavior (HA-CoPB) Framework: This proposed framework serves as the foundation for persona modeling, ensuring that the behavioral aspects of travel decisions are accurately represented.
- Structured Multi-Agent Conversation: PEMANT engages in a two-phase conversation model to simulate real-world household decision negotiations, enhancing the quality of trip planning outcomes.
- Persona-Alignment Control Mechanism: A novel mechanism that ensures personas remain aligned throughout the negotiation process, improving the realism of interactions among agents.
Evaluation and Results
PEMANT has been rigorously evaluated using both national and regional household travel survey datasets. The results demonstrate that PEMANT consistently outperforms state-of-the-art benchmarks across various datasets, showcasing its effectiveness in predicting travel behavior accurately.
Conclusion
In summary, PEMANT represents a significant advancement in the field of travel demand modeling by integrating behavioral theory and LLM capabilities. This framework not only enhances the predictive accuracy of household-level trip generation but also facilitates a deeper understanding of the dynamics involved in collective travel decisions. As urban planning continues to evolve, frameworks like PEMANT could play a crucial role in shaping efficient and sustainable transport systems.