HyDE
A RAG retrieval optimization technique in which a language model first generates a hypothetical answer text, and the vector representation of that text is then used to retrieve real documents. This approach reduces the semantic gap between the original query and the target documents, improving retrieval relevance.
Core Facts
Timeline (last 90 days)
Query expansion, HyDE, and spelling normalization are often more effective on the query side than swapping the embedding model.
SAFE-QE injects only 4.3 unsupported terms per query, compared to 27.8 for Query2doc and 57.5 for HyDE.
Query2doc achieves LLM-based query expansion by generating short paragraphs, while HyDE does so by generating hypothetical documents.
The HyDE method injects approximately 13 times more hallucinated terms per query than SAFE-QE.
The cost of HyDE is one additional LLM call per retrieval, making it suitable for scenarios that are insensitive to latency but have high recall requirements.
The idea of hierarchical summary indexing has re-emerged in mainstream RAG frameworks in the form of variants such as HyDE and parent-child document retrieval.
HyDE (Hypothetical Document Embeddings) has the model first generate a hypothetical answer, then uses the vector of that text to retrieve real documents, in order to close the semantic gap between the query and the documents.
The first stage uses HyDE-style retrieval, but the query is not the original clause itself; instead, it is a hypothetical, mutually contradictory version of that clause.
Advanced RAG introduces Query Rewriting and HyDE (Hypothetical Document Embeddings) before retrieval to optimize query quality, and improves relevance after retrieval through Cross-Encoder re-ranking and context compression.
All Facts (10)
HyDE (Hypothetical Document Embeddings) is a query rewriting strategy in which a large model first generates a hypothetical answer document, which is then used for retrieval.
75%UnverifiedQuery expansion, HyDE, and spelling normalization are often more effective on the query side than swapping the embedding model.
50%UnverifiedQuery2doc achieves LLM-based query expansion by generating short paragraphs, while HyDE does so by generating hypothetical documents.
50%UnverifiedSAFE-QE injects only 4.3 unsupported terms per query, compared to 27.8 for Query2doc and 57.5 for HyDE.
50%UnverifiedThe HyDE method injects approximately 13 times more hallucinated terms per query than SAFE-QE.
50%UnverifiedThe cost of HyDE is one additional LLM call per retrieval, making it suitable for scenarios that are insensitive to latency but have high recall requirements.
50%UnverifiedThe idea of hierarchical summary indexing has re-emerged in mainstream RAG frameworks in the form of variants such as HyDE and parent-child document retrieval.
50%UnverifiedHyDE (Hypothetical Document Embeddings) has the model first generate a hypothetical answer, then uses the vector of that text to retrieve real documents, in order to close the semantic gap between the query and the documents.
50%UnverifiedThe first stage uses HyDE-style retrieval, but the query is not the original clause itself; instead, it is a hypothetical, mutually contradictory version of that clause.
50%UnverifiedAdvanced RAG introduces Query Rewriting and HyDE (Hypothetical Document Embeddings) before retrieval to optimize query quality, and improves relevance after retrieval through Cross-Encoder re-ranking and context compression.
50%