GLOSSARY

Hybrid Search

Hybrid search combines keyword and vector retrieval. It keeps precision while adding semantic recall.

What is Hybrid Search?

Hybrid search merges lexical (BM25/phrases) and vector signals in recall or ranking. It protects exact brands/SKUs while rescuing paraphrased queries.

How It Works (quick)

  • Dual recall: Run BM25 and vector recall in parallel.
  • Merge: De-duplicate and union candidates.
  • Re-rank: Score with features (lexical, vector, business) via LTR/neural.
  • Guardrails: Hard filters and diversity caps.

Why It Matters in E-commerce

  • Best of both: Exact matches + semantic coverage.
  • Reduces zero-results without noisy expansions.

Best Practices

  • Keep exact fields (brand/SKU) prioritized.
  • Calibrate per category/locale.
  • Cap vector influence; log contributions.

Summary

Hybrid search is the pragmatic default: lexical precision plus semantic reach, under clear caps.