Semantic Retrieval

Semantic Retrieval is a search method based on embeddings and meaning understanding that matches documents based on concepts, not exact words. It solves the Vocabulary Mismatch problem — understanding that 'pool' = 'swimming pool' = 'aquapark'. In Google's hybrid system, semantic retrieval works simultaneously with lexical retrieval, and their results are combined during reranking.

This is the foundation for many applications: internal linking (nearest neighbors), duplicate detection, content clustering, and calculating Site Focus Score. Embeddings (numerical vectors representing meaning) are the technology behind semantic retrieval.

In practice, to use semantic retrieval to your advantage, cover an entity's semantic field — use synonyms, hyponyms, meronyms, and related concepts. Lexical expansion tools generate these extensions automatically, increasing the number of touchpoints with queries.