What is Query Expansion? 💡

Query Expansion is the process of enhancing a user’s original search query by adding synonyms or related terms. This makes the query more comprehensive, allowing the search engine to retrieve more relevant results.

📌 In Our System:

• 🔹 Closely linked to Statistical Machine Translation.
• 🔹 Driven by a context map storing synonyms, usage contexts, and relevance scores. ✅

📊🗺️ The Role of the Context Map & Synonym Scoring

What Is the Context Map?

The context map (580) is a data structure that contains words from past queries along with potential synonyms.

• 🔹 Associated Context: Information about words appearing before and after the original word.
• 🔹 A Score: A numerical value representing the likelihood that a synonym is appropriate in a given context.

How Is the Score Determined?

The score is derived from the translation likelihood produced by the machine translation model, which includes:

• 🌍 Language Probability: How natural the output is in the target language.
• 🔀 Translation Probability: The likelihood that the output text is a valid translation of the input.

Example:

For the word “tie”, the context map might list:

• 🪢 “knot” (high score, fits well in “how to tie a”)
• 👔 “windsor” (lower score, more relevant in “tie a windsor knot”)

🔎 The Query Expansion Process

Step 5: Match Contexts and Select the Best Synonym (Step 650)

• 🔹 Context Matching: The system compares the word’s context in the original query to those in the context map.
• 🔹 Left Context: Example: “how to tie a bow” → “how to tie a”
• 🔹 Right Context: Example: “how to tie a bow” → “a bow”

The synonym with the highest score is selected.

Step 6: Expand the Query with the Selected Synonym (Step 660)

• 🔹 Appending: Simply add the synonym to the query.
• 🔹 Reformulation: Use logical OR to include synonyms.

Example: “how to tie a bow” → “how to (tie OR knot) a bow”

Step 7: Use the Expanded Query to Search the Corpus (Step 670)

The expanded query retrieves more comprehensive and relevant results by including synonyms.

📌 On-Line vs. Off-Line Translation for Query Expansion

🖥️ On-Line (Synchronous) Translation

The search query is translated on the fly.

• 🔹 How It Works: The system identifies synonyms by comparing original and translated queries.
• 🔹 Example: “how to become a mason” → “how to be a bricklayer” (revealing synonyms).

📁 Off-Line (Asynchronous) Translation

Translations for batches of queries are pre-computed and stored.

• 🔹 How It Works: New queries use pre-existing translations for expansion.
• 🔹 Benefit: More efficient for resource-intensive translation tasks.

🏆 Real-World Example: Expanding a Query

Example Query:

Original Query: “how to tie a bow”

• 🔹 Word Selected for Expansion: “tie”
• 🔹 Context Map Lookup: “knot” (high score), “windsor” (lower score).
• 🔹 Reformulated Query: “how to (tie OR knot) a bow”
• 🔹 Outcome: Improved search results including both terms.