Query Transformation

What is Query Transformation?

Query transformation is the set of rewrites and enrichments applied to a raw query before retrieval and ranking. It includes normalization, spell correction, synonym expansion, phrase/entity detection, unit/currency conversion, and operator parsing (e.g., “under €150” → price ≤ 150).

How It Works (quick)

• Normalize: Lowercase, accent fold, trim punctuation; unify hyphen/space variants.
• Correct: Length-aware spelling/keyboard fixes (skip exact fields like SKU/brand).
• Expand (late-bound): Category- and locale-aware synonyms (trainers ↔ sneakers), abbreviations (GTX ↔ GORE-TEX).
• Understand: Extract entities/constraints (size, color, price, date), detect phrases, convert units/currency.
• Rewrite: Build a structured plan: filters/facets, keyword fields, phrase/bigram queries.
• Guardrails: Caps on expansions, allow undo/preview, log what fired for audits.

Why It Matters in E-commerce

• Higher recall, same precision: Recover results for typos and vocabulary gaps without flooding noise.
• Fewer steps: One natural query can apply multiple filters instantly.
• Localization: Handles sizes, units, and slang per market.

Best Practices

• Prefer late binding for synonyms and boosts; keep index lean.
• Protect exact fields (SKU/MPN/brand) from fuzz and aggressive rewrites.
• Use context (category, user locale) to choose expansions; avoid global rules.
• Render chips for extracted constraints so users can edit quickly.
• A/B test rewrite policies and monitor zero-results, CTR, reformulations.
• Version configs and keep explain logs for each transformation.

Challenges

• Over-expansion causing off-topic hits; multilingual ambiguity; latency from multiple steps; privacy with personalization features.

Examples

• “nike gore tex trail shoes size 45 under 150”
• → synonyms: gore tex → GORE-TEX; entities: size=45, price ≤ €150; filters + phrase field for trail shoes.
• “AB-1234”
• → bypass transforms; route to exact SKU field only.
• “laptop 16gb 512 ssd”
• → entities: RAM=16 GB, SSD=512 GB; map to attributes and phrase/bigram fields.

Summary

Query transformation turns messy input into a structured, context-aware query plan. Keep rewrites late-bound, safe, and explainable to boost recall and speed without sacrificing precision.