HyDE
Hypothetical Document Embeddings is a retrieval technique where an LLM generates a fake ideal document to answer a query, which is then vectorized.
Think of It Like This
Like sketching a picture of a suspect based on a description, and then using that sketch to search a photo database.
Instead of matching a user query directly to documents, HyDE generates a hypothetical response and uses its embedding to search the vector database. This maps the search into the document's semantic space rather than the query's space. It is highly effective for zero-shot retrieval tasks without requiring relevance training data.