Bridging the Cold-Start Gap: LLM-Powered Synthetic Data Generation for Natural Language Search at Airbnb

A batch of recent arXiv papers (May 22, 2026) focused on applying and improving large language models (LLMs) for search, recommendation, retrieval, and related ML infrastructure. Key contributions include: LLM-powered synthetic query and label generation to solve cold-start for Airbnb natural-language search; reinforced preference optimization (RPORec) to align LLM reasoning with recommendation heads; LLM-based semantic candidate generation to improve stability/predictability for ads recommendation; a fully generative conversational recommender unifying item prediction and dialog; behavior‑dr