Researchers have introduced Behavior2Trip (B2T), a new benchmark and agent designed for personalized travel planning by analyzing user behavior trajectories. Unlike existing methods that rely on explicit instructions, B2T infers preferences from past actions, reducing user interaction burden. The benchmark, derived from a large Chinese travel platform, contains 11,400 instances of user behaviors. Experiments show that current large language models like GPT-4.1 struggle with this task, achieving low pass rates, while the proposed B2T-Agent, particularly when built on Qwen3-8B, demonstrates superior performance and generalization capabilities. AI
IMPACT This research highlights the challenge of inferring user preferences from behavior, potentially leading to more sophisticated personalized AI agents in various domains.
RANK_REASON The cluster describes a new academic paper introducing a novel benchmark and agent for a specific AI task. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
