Mapping Search Queries To Search Intents
Source: https://medium.com/@dtunkelang/search-queries-and-search-intent-1dec79ad155f
Author: Daniel Tunkelang
Summary
Daniel Tunkelang examines the relationship between search queries (what users type) and search intents (what users want), arguing that this mapping is the core challenge of Query Understanding and that most search failures are failures of intent mapping, not retrieval.
The Query-Intent Gap
Users rarely express their full intent in a query. A user searching for “python” on a developer site has a completely different intent than the same query on a nature website.
The gap:
- Query: short, ambiguous, underspecified
- Intent: rich, contextual, often multi-faceted
Closing this gap = query understanding.
Intent Dimensions
Daniel Tunkelang identifies multiple dimensions of intent beyond the simple navigational/informational/transactional taxonomy:
1. Topical Intent
What subject area? “python” → programming OR reptile
2. Task Intent
What task? Informational (learn about python) vs. transactional (download python) vs. navigational (go to python.org)
3. Specificity Intent
How specific? “shoes” (browse) vs. “Nike Air Max 90 size 10.5 white” (exact match)
4. Quality Intent
What quality matters? Price-sensitive, brand-loyal, novelty-seeking
5. Freshness Intent
How recent? “latest iphone” vs. “best iphone for older users” (historical)
Intent Signals in Queries
Linguistic patterns that signal intent:
| Signal | Example | Intent Implied |
|---|---|---|
| Question words | ”how to…”, “what is…” | Informational |
| Buy/price words | ”buy”, “price”, “cheap” | Transactional |
| Site-like queries | ”amazon”, “facebook login” | Navigational |
| Comparison words | ”vs”, “compare”, “best” | Comparative |
| Vague adjectives | ”cool”, “nice”, “interesting” | Discovery |
The “Intent Not Inventory” Principle
Tunkelang’s key insight (later developed in “Search: Intent, Not Inventory”):
A search for “birthday cake ideas” is not a search for documents containing those words. It’s a search for inspiration, recipes, images, and perhaps local bakeries.
The search system should serve the intent (birthday party planning help), not literally match the query (documents about birthday cake ideas).
Practical Implications
- For ambiguous queries: Diversity Metrics to cover multiple intents
- For high-confidence intent: streamline to directly serve the intent
- For transactional intents: structured data (price, availability) matters as much as text
- For discovery intents: personalization and trending signals matter
Related Articles
- Search - Intent Not Inventory — same author, extends this argument
- Query Understanding - Divided into Three Parts — three-part framework includes intent
- AI for Query Understanding — LLM approach to intent classification
- Food Discovery with Uber Eats — industrial intent mapping at scale
Related Concepts
- Search Intent — primary concept
- Query Understanding — framework containing intent mapping
- Query Types — how types map to intents
- Diversity Metrics — for multi-intent queries
People
- Daniel Tunkelang — author