Abstract:As global tourism expands and artificial intelligence technology advances, intelligent travel planning services have emerged as a significant research focus. Within dynamic real-world travel scenarios with multi-dimensional constraints, services that support users in automatically creating practical and customized travel itineraries must address three key objectives: Rationality, Comprehensiveness, and Personalization. However, existing systems with rule-based combinations or LLM-based planning methods struggle to fully satisfy these criteria. To overcome the challenges, we introduce TravelAgent, a travel planning system powered by large language models (LLMs) designed to provide reasonable, comprehensive, and personalized travel itineraries grounded in dynamic scenarios. TravelAgent comprises four modules: Tool-usage, Recommendation, Planning, and Memory Module. We evaluate TravelAgent's performance with human and simulated users, demonstrating its overall effectiveness in three criteria and confirming the accuracy of personalized recommendations.
What problem does this paper attempt to address?
### Problems the Paper Attempts to Solve
With the development of global tourism and the advancement of artificial intelligence technology, intelligent travel planning services have become a significant research focus. In dynamic real-world travel scenarios, services that support users in automatically generating practical and personalized travel plans need to meet three key objectives: Rationality, Comprehensiveness, and Personalization. However, existing planning methods based on rule combinations or large language models (LLMs) find it challenging to fully meet these standards.
To overcome these challenges, this paper introduces **TravelAgent**, a travel planning system driven by large language models, designed to provide rational, comprehensive, and personalized travel plans that adapt to dynamic scenarios. TravelAgent comprises four modules: Tool-usage Module, Recommendation Module, Planning Module, and Memory Module. Evaluations by human users and simulated users demonstrate the overall effectiveness of TravelAgent on these three standards and confirm the accuracy of personalized recommendations.
### Specific Problems and Solutions
1. **Rationality**:
- How to model constraints in travel scenarios and formulate reasonable itineraries under these constraints?
- Enhance the rationality of the planning phase by modeling travel constraints and developing spatiotemporal-aware planning algorithms.
2. **Comprehensiveness**:
- How to provide comprehensive travel services, including recommendations and planning, to generate real-time, detailed, and engaging itineraries?
- Establish a real-time data mechanism to enhance the functionality of multiple tools, ensuring the comprehensiveness and practicality of the itinerary.
3. **Personalization**:
- How to discover and utilize implicit user personalization information to provide personalized recommendation and planning services?
- Introduce an agent-based personalized recommendation framework to enhance personalization through continuous learning of implicit information.
### Main Contributions
- **System Design**: Defined the three objective standards of travel services and created TravelAgent, a novel AI-driven system for automated personalized travel planning services.
- **Method Innovation**: For rationality, modeled travel constraints and developed spatiotemporal-aware planning algorithms; for comprehensiveness, established a real-time data mechanism and enhanced the functionality of multiple tools; for personalization, introduced an agent-based personalized recommendation framework to enhance personalization through continuous learning of implicit information.
- **Evaluation Methods**: Evaluated the overall performance of TravelAgent through human case studies, surpassing the baseline GPT-4+ agent in the three standards. Additionally, through behavior prediction tasks with simulated users, found that our personalized recommendation framework has a lower error rate compared to the baseline LLM-based recommendation methods.
### Conclusion
This paper introduces a new travel service system, **TravelAgent**, which utilizes large language models and real-time data to automatically generate rational, comprehensive, and personalized travel plans in real-world scenarios with multidimensional constraints. Evaluations by human users and simulated users validate the effectiveness of TravelAgent in terms of rationality, comprehensiveness, and personalization.