Query Parsing

Query parsing is the process of breaking down a search query into its components to understand its meaning and intent. This helps search engines or systems deliver relevant results.

📌 1. Types of Queries

• 🔹 Entity-Seeking Query
  Searches for a specific thing (person, place, product, etc.).
  ✅ Example:
  "What hotel looks like a sail" → The user wants the exact price of a specific product.
  "Who invented the telephone?" → The user seeks a factual answer about a historical figure.
• 🔹 Substitute Term
  Words that replace another term but aren’t exact synonyms.
  ✅ Example:
  "Fix laptop screen" vs. "Repair notebook display"
  "Notebook" substitutes "laptop," and "display" substitutes "screen."
• 🔹 Synonym Term
  Words with the same meaning used interchangeably.
  ✅ Example:
  "Buy sneakers" vs. "Purchase athletic shoes"
  "Buy" = "Purchase"
  "Sneakers" = "Athletic shoes"
• 🔹 Canonical Query
  The "standard" version of a query that covers variations.
  ✅ Example:
  Canonical: "How to lose weight"
  Variations:
  "Best ways to shed pounds"
  "Tips for weight loss"
  💡 Search engines treat these as the same intent despite wording differences.

🔸 Additional Query Types

• 1️⃣ 𝐂𝐚𝐧𝐨𝐧𝐢𝐜𝐚𝐥 𝐐𝐮𝐞𝐫𝐲: is the main version of a search query that represents the core intent behind different variations of searches.
• 2️⃣ 𝐑𝐞𝐩𝐫𝐞𝐬𝐞𝐧𝐭𝐚𝐭𝐢𝐯𝐞 𝐐𝐮𝐞𝐫𝐲: is a broader version of query after canonical query.
• 3️⃣ 𝐑𝐞𝐩𝐫𝐞𝐬𝐞𝐧𝐭𝐞𝐝 𝐐𝐮𝐞𝐫𝐢𝐞𝐬: The different ways people search for the same thing.
  For example:
  🔸 Canonical Query: "Flights to Dubai" (broad and standard version that covers all related searches).
  🔸 Representative Query: "cheap flights to Dubai" (chosen as the most relevant version).
  🔸 Represented Queries: "cheap flights to Dubai," "budget airfare to Dubai," "low-cost Dubai flights."
• 4️⃣ 𝐒𝐮𝐛𝐬𝐭𝐢𝐭𝐮𝐭𝐞 𝐐𝐮𝐞𝐫𝐲: is an alternative version of a search query that Google might use to improve search results. It helps when a user’s original query isn’t clear or could be phrased better.
  Example:
  User searches: "cheap hotels near me"
  Google detects that "affordable hotels nearby" is a close match using co-occurrence and vector analysis.
• 5️⃣ 𝐀𝐧𝐬𝐰𝐞𝐫-𝐒𝐞𝐞𝐤𝐢𝐧𝐠 𝐐𝐮𝐞𝐫𝐲: is a search query where the user expects a clear, direct answer rather than just a list of web pages.
  Examples:
  “Who is the CEO of Tesla?” → Direct Answer: Elon Musk
  “How to tie a tie?” → Featured Snippet with step-by-step instructions
• 6️⃣ 𝐍𝐚𝐭𝐮𝐫𝐚𝐥 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐐𝐮𝐞𝐫𝐲 (𝐍𝐋𝐐): is a search query written in a conversational or human-like way, often without a single factual answer.
  Example:
  “How do I make hummus?”
  Variations exist (classic, spicy, beet hummus).
  Includes "I" → Personalized intent.
  Multiple Possible Answers → Different methods, opinions, or interpretations.
• 7️⃣ 𝐒𝐞𝐞𝐝 𝐐𝐮𝐞𝐫𝐲: is a starting point for search engines to understand different variations of a query and user intent.
  These queries can be either:
  • 8️⃣ 𝐒𝐲𝐧𝐭𝐡𝐞𝐭𝐢𝐜 𝐐𝐮𝐞𝐫𝐢𝐞𝐬 → AI-generated queries based on patterns.
    (Here comes the concept of 𝐐𝐮𝐞𝐫𝐲 𝐑𝐞𝐰𝐫𝐢𝐭𝐢𝐧𝐠: Search engines rewrite a query to better understand the context when a query is ambiguous or unclear)
  • 9️⃣ 𝐔𝐬𝐞𝐫-𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐞𝐝 𝐐𝐮𝐞𝐫𝐢𝐞𝐬 → Real queries typed by users.
• 𝐖𝐡𝐚𝐭 𝐢𝐬 𝐐𝐮𝐞𝐫𝐲 𝐁𝐫𝐞𝐚𝐝𝐭𝐡?
  Refers to how broad or narrow a search query is based on adjacent words and unknown entities.
  If many documents match the query → 𝐇𝐢𝐠𝐡 𝐐𝐮𝐞𝐫𝐲 𝐁𝐫𝐞𝐚𝐝𝐭𝐡 (broad topic).
  If few documents match → 𝐋𝐨𝐰 𝐐𝐮𝐞𝐫𝐲 𝐁𝐫𝐞𝐚𝐝𝐭𝐡 (specific topic).

📌 3. Query Character & SERP Design

• The query’s "character" determines how results are displayed on the SERP (Search Engine Results Page).
  Example 1:
  Query: "Facebook login" (🔗 Navigational Intent)
  SERP: Direct link to Facebook’s login page.
• Example 2:
  Query: "Is caffeine bad for health?" (📖 Informational Intent)
  SERP: Articles, studies, and featured snippets explaining pros/cons.
• 🎯 Dominant vs. Minor Intent
  Example:
  Query: "Starbucks near me open now"
  Dominant Intent: Find a nearby Starbucks that’s currently open.
  Minor Intent: Maybe check reviews or menu (secondary info).

📌 4. Query Process (Steps in Parsing)

• 🔹 Step 1: Tokenization → Split the query into words/phrases.
  "Best action movies 2023" → ["best", "action", "movies", "2023"]
• 🔹 Step 2: Identify synonyms/substitutes → Replace "action" with "thriller" if needed.
• 🔹 Step 3: Classify intent → Is this seeking a list? Reviews? Comparisons?
• 🔹 Step 4: Map to canonical form → Standardize to "top action films 2023".