Natural language understanding (NLU) computationally interprets human language so search systems can identify meaning, entities, concepts, relationships, and intent. It applies to both indexed content and user queries, helping a system move beyond literal keyword matching toward more relevant answers. section 2.0
Search-oriented NLU works across several layers of language:
- Recognize keywords, entities, concepts, misspellings, synonyms, acronyms, and ambiguous terms. section 2.0
- Model explicit and implied relationships, taxonomies, ontologies, and knowledge graphs. section 2.0
- Use distributional semantics and vector-based embeddings to derive and score semantic relationships between query and document term sequences. section 2.5
- Combine content with behavioral signals, such as user search activity, to support learning and relevance improvement. section 2.6
From query terms to intent
A search for an ambiguous term may require entity recognition, synonym handling, relationship modeling, and semantic comparison with document content. User signals can then provide additional evidence for learning which results best satisfy the query. section 2.0 section 2.6
Large language models provide strong general-knowledge interpretation of meaning and intent, while domain-specific applications may require fine-tuning. They are powerful but remain one part of a broader search toolbox that includes knowledge modeling, behavioral interpretation, and relevance ranking. section 2.6
