Sentence Transformers Semantic Search Documentation

Embedding-Based Retrieval Mechanism

The documentation explains that semantic search improves accuracy by understanding meaning rather than relying on lexical matches. It handles synonyms and misspellings better than keyword engines. The core mechanism involves embedding corpus entries into a vector space. At search time, the query is embedded into the same space to find closest matches based on semantic similarity.

Symmetric and Asymmetric Distinctions

The source distinguishes between symmetric and asymmetric semantic search setups. Symmetric search applies when queries and corpus entries have similar length and content volume. An example provided is searching for similar questions. This distinction is critical for configuring the system correctly. The documentation does not provide specific performance metrics or hardware compatibility details for local deployment scenarios.