Meituan Search 3.0: Exploring and Applying LLM Semantic Representations in Ranking Models

Under the profound influence of Large Language Model (LLM) technology, search engines are undergoing a third paradigm shift: evolving from "keyword text matching" in the 1.0 era, to "behavioral statistics and personalized search" in the 2.0 era, and now towards comprehensive evolution toward "complex intent understanding and cognitive decision-making" in the 3.0 era. The Meituan Search team is leveraging its new architecture for integrating general and vertical capabilities with generative LLM technologies to comprehensively reconstruct the foundation of local life search. This series of technical blogs will continue to introduce Meituan Search 3.0's technical explorations; this article focuses on three phases of practice in applying LLM semantic representations to service retail ranking scenarios—from single-point feature validation to building a systematic representation framework, and finally to cross-scenario transfer and reuse—exploring the application path of semantic matching signals in search ranking. Under the profound influence of Large Language Model (LLM) technology, search engines are undergoing a third paradigm shift: evolving from "keyword text matching" in the 1.0 era, to "behavioral statistics and personalized search" in the 2.0 era, and now towards comprehensive evolution toward "complex intent understanding and cognitive decision-making" in the 3.0 era. The Meituan Search team is leveraging its new…