Knowledge
Knowledge HubConcepts and strategic researchGlossaryDefinitions across six domainsFrameworksStructured strategy methodsTopicsConnected subject areasLibrary
PlaybooksPractical strategy guidesBooksBooks and long-form researchOperator VaultMember resources and toolkits
Explore Discover, the Knowledge Hub, playbooks, and the Operator Vault.
Glossary
Semantic search
Semantic search is the retrieval objective of finding information by meaning and relatedness rather than by exact keyword match alone. Implementations vary: many contemporary systems use dense embedding retrieval, hybrid lexical–dense ranking, knowledge graphs, or query expansion — dense retrieval is a common technique, not the whole field.
Why it matters
Keyword search fails when users and documents use different words for the same intent. Semantic search improves recall on paraphrases and conceptual queries, which is why it underpins many enterprise knowledge assistants and RAG pipelines.
Dense retrieval (embedding queries and documents, then ranking by vector similarity) is one widely used implementation. “Neural search” is informal product language that often overlaps with dense or learned ranking methods. Neither phrase should be treated as a strict synonym for semantic search as a whole.
Example (enterprise): Engineers search “how do we rotate service credentials?” and retrieve runbooks titled “secret lifecycle” and “key vault rotation” because the retrieval stack matches intent, even though the query terms never appear — typically via dense or hybrid methods evaluated against keyword baselines.
Mechanisms
- Dense retrieval — embed queries and documents; rank by vector similarity (one major implementation path).
- Hybrid retrieval — combine dense scores with lexical scores (for example BM25) to catch exact identifiers and rare tokens.
- Re-ranking — optional cross-encoders or LLM re-rankers improve precision on the top candidates.
Boundaries
- Semantic search ≠ dense retrieval as exact synonyms; dense retrieval is a technique that often serves the semantic-search objective.
- Semantic search ≠ RAG. RAG uses retrieval (often semantic) to condition generation; search can stop at ranked documents.
- Semantic search ≠ vector database. The database is storage/index infrastructure.
- Semantic matches can still be wrong for the user’s task; measure relevance with domain judgments, not only vector distance.
Related topics
None linked yet.
Related concepts & frameworks
Further research
No related resources yet.