Year: 2026
Type: conference-paper
Abstract: Travel itinerary generation requires balancing strict spatio-temporal constraints with human preferences.Existing LLM-based planners mainly rely on structured attributes and predefined traveler personas, but real travel decisions are often shaped by reviews that reveal experiential factors such as comfort, safety, service quality, ambiance, crowding, and hidden risks absent from structured databases.Incorporating such review information is therefore critical to realistic, user-centric itinerary generation.We propose TRIPPULSE 1 , a multiagent framework for review-grounded travel planning.Instead of relying on a monolithic planner (and face context and reasoning bottlenecks), TRIPPULSE 2 decomposes itinerary generation into specialized agents (each operating over localized contexts) for accommodations, transportation, meals, attractions, and events, coordinated through a global orchestrator with scheduling mechanisms that enforce temporal and budget feasibility.We augment TRIPCRAFT with 100K+ real-world reviews and introduce Review-Grounded Persona Alignment (RGPA), an LLM-as-a-Judge metric for evaluating alignment with humancentric travel experiences.Experiments across multiple trip durations and diverse proprietary and open-source models show that TRIPPULSE maintains strong constraint satisfaction while generating more personalized and experientially grounded itineraries.
Institutions: Indian Institute of Technology Bhubaneswar, Microsoft (India)
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7204972067