What is a Search Algorithm?
A search algorithm is the set of retrieval and ranking steps that turn a query into an ordered list of results. It combines lexical match (BM25/phrases), semantic signals (vectors), filters/ACL, and business features (stock, rating, price, margin).
How It Works (quick)
- Analyze query: Language detect → tokenize/lemmatize → detect phrases/entities.
- Recall: Lexical (inverted index), optional dense retrieval (ANN vectors).
- Filter: Apply hard rules first (OOS, region, permissions).
- Rank: Blend text scores with quality/business features; optional Learning to Rank.
- Controls: Diversity, brand caps, promotions; log why this result.
Why It Matters in E-commerce
- Speed to product: Clean top-k reduces pogo-sticking.
- Business fit: Surfaces in-stock, high-quality items that match intent.
- Explainability: Highlights and score breakdowns build trust.
Best Practices
- Hybrid retrieval: BM25 + vectors; keep exact fields for SKU/brand.
- Category/locale tuning: Footwear ≠ electronics; en-GB ≠ en-US.
- Guardrails: Cap popularity/freshness boosts; enforce OOS and compliance early.
- Observability: Track NDCG/CTR/conv, zero-results, tail latency; keep golden sets.
- Versioning: Configs and models with rollback.
Summary
A search algorithm is retrieval + ranking under guardrails. Hybrid evidence plus clear controls delivers relevant, shoppable results.
