Hybrid Search
Combining keyword and vector retrieval, then merging the two rankings.
Überblick
Hybrid search runs a lexical and a semantic query in parallel and fuses the results, usually by reciprocal rank. It solves the problem that each method fails differently: keyword search misses paraphrases, and embedding search misses exact identifiers, names and rare tokens. It exists because the union of the two is more robust than either, which is what most production retrieval actually runs.
Technologiegraph
Wie Hybrid Search mit dem übrigen Stack zusammenhängt.
Nutzt und hängt ab von
- Embedding concept Techniques Hybrid Search uses Embedding High ✓ source
Angewendet auf
- AI Search application Applications Hybrid Search used for AI Search High ✓ source