Search Intent

Definition

Search intent describes what a user is actually trying to accomplish with a query — distinct from what words they typed. Understanding intent allows search systems to serve the user’s underlying need rather than just matching keywords.

Classic Taxonomy (Broder 2002)

Navigational Intent

User wants to go to a specific website or page.

• “Facebook login”
• “New York Times homepage”
• Best served by: direct link, no ranked list needed

Informational Intent

User wants to learn something.

• “How does photosynthesis work?”
• “symptoms of appendicitis”
• Best served by: comprehensive, authoritative content

Transactional Intent

User wants to do something — often buy, download, or sign up.

• “buy Nike running shoes”
• “download Spotify”
• Best served by: product pages, conversion-optimized results

Extended Taxonomy for E-commerce

Intent	Query Example	Signal	Optimal Response
Navigational	”my orders”	Exact brand/URL	Direct page
Informational	”how to clean leather shoes”	Question words	Guide/article
Transactional	”buy red sneakers size 10”	Buy/purchase words	Product listing
Comparative	”Nike vs Adidas running shoes”	vs, compare	Comparison page
Discovery	”cool summer outfits”	Vague, exploratory	Curated collection

Intent Classification

Modern intent classifiers use fine-tuned BERT/RoBERTa:

from transformers import pipeline
 
classifier = pipeline("text-classification", 
                       model="domain/intent-classifier")
result = classifier("buy cheap running shoes")
# → {"label": "transactional", "score": 0.94}

Features that signal intent:

• Question words (what, how, why) → informational
• Action verbs (buy, download, sign up) → transactional
• Brand names without product → navigational
• Adjectives + category → discovery/exploratory

Intent-Retrieval Alignment

Different intents need different retrieval strategies:

Intent	Retrieval Strategy
Navigational	Exact match / URL lookup
Informational	Semantic Search / RAG
Transactional	Structured data + Hybrid Search
Discovery	Diversity Metrics + personalization

Ambiguous Queries

Many queries are intent-ambiguous:

• “python”: programming vs. snake
• “mercury”: planet, element, car, Roman god, Freddie Mercury

For ambiguous queries, Diversity Metrics / MMR ensures coverage of multiple intents.

Relation to Agentic Search

In Agentic Search, the agent must infer intent to choose the right retrieval strategy:

• Informational → comprehensive RAG
• Transactional → structured search + ranking
• Comparative → multi-source retrieval + synthesis

Related Concepts

• Query Understanding — intent is a component of QU

• Query Types — more granular taxonomy

• Diversity Metrics — for ambiguous intent

• Agentic Search — intent-aware retrieval planning

• Bag-of-Documents Model — probabilistic intent framing

• Intent Drift — how the right answer for a query shifts over time

People

• Daniel Tunkelang — “Mapping Search Queries To Search Intents”, “Search: Intent, Not Inventory”