Trip+: Benchmarking Agents in Personalized Interactive Travel Planning

The article introduces Trip+, a new benchmark designed to evaluate the performance of language models in personalized interactive travel planning. Trip+ assesses agents' abilities to generate and revise detailed itineraries based on traveler profiles and dynamic interactions, using an LLM-based simulator to measure subjective metrics such as traveler fatigue. The evaluation of 18 language models revealed a significant gap in experiential quality, with models often producing itineraries that are technically feasible but misaligned with user preferences, highlighting the need for improved personalization in AI-driven travel planning applications.