Why relevance scoring ranks results in Elasticsearch - Visual Breakdown

Concept Flow - Why relevance scoring ranks results

User Query

↓

Search Engine

↓

Match Documents

↓

Calculate Relevance Scores

↓

Sort Documents by Score

↓

Return Ranked Results

The search engine takes a user query, finds matching documents, calculates a relevance score for each, sorts them by score, and returns the ranked list.

Execution Sample

Elasticsearch

GET /my_index/_search
{
  "query": {
    "match": { "content": "apple" }
  }
}

This query searches for documents containing 'apple' and ranks them by relevance score.

Execution Table

Step	Action	Documents Matched	Score Calculation	Sorted Order
1	Receive query 'apple'	-	-	-
2	Find documents containing 'apple'	Doc1, Doc2, Doc3	-	-
3	Calculate score for Doc1	Doc1	TF=3, IDF=1.5, Score=4.5	-
4	Calculate score for Doc2	Doc2	TF=1, IDF=1.5, Score=1.5	-
5	Calculate score for Doc3	Doc3	TF=2, IDF=1.5, Score=3.0	-
6	Sort documents by score	Doc1, Doc3, Doc2	-	Doc1 > Doc3 > Doc2
7	Return ranked results	Doc1, Doc3, Doc2	-	-

💡 All matched documents scored and sorted; results returned in descending order of relevance.

Variable Tracker

Variable	Start	After Step 3	After Step 4	After Step 5	Final
Doc1_Score	0	4.5	4.5	4.5	4.5
Doc2_Score	0	0	1.5	1.5	1.5
Doc3_Score	0	0	0	3.0	3.0

Key Moments - 3 Insights

Why does Doc1 have a higher score than Doc3 even though both contain 'apple'?

Doc1 has a higher term frequency (TF=3) compared to Doc3 (TF=2), so its score is higher as shown in steps 3 and 5 of the execution_table.

What does the IDF value represent in the score calculation?

IDF (Inverse Document Frequency) measures how rare the term 'apple' is across all documents, boosting scores for rarer terms; it is constant (1.5) for all docs here as shown in score calculations.

Why are documents sorted after scoring?

Sorting by score ensures the most relevant documents appear first in results, as shown in step 6 where Doc1 > Doc3 > Doc2.

Visual Quiz - 3 Questions

Look at the execution_table, what is the score of Doc3 after step 5?

A 4.5

B 1.5

C 3.0

D 0

💡 Hint

Check the 'Score Calculation' column at step 5 in the execution_table.

At which step are the documents sorted by their relevance score?

A Step 3

B Step 6

C Step 2

D Step 7

💡 Hint

Look for the 'Sort documents by score' action in the execution_table.

If Doc2 had TF=4 instead of 1, how would its score change after step 4?

A Score would be 6.0

B Score would be 1.5

C Score would be 4.5

D Score would be 3.0

💡 Hint

Score = TF * IDF; IDF is 1.5 as shown in the execution_table.

Concept Snapshot

Relevance scoring ranks search results by calculating a score for each document.
Score depends on term frequency (TF) and inverse document frequency (IDF).
Higher scores mean more relevant documents.
Results are sorted by score before returning.
This helps users see the best matches first.

Full Transcript

When a user searches, Elasticsearch finds documents containing the query terms. Each document gets a relevance score based on how often the term appears (TF) and how rare the term is across all documents (IDF). These scores are calculated step-by-step for each document. Then, documents are sorted by their scores so the most relevant appear first. For example, Doc1 with TF=3 scores higher than Doc3 with TF=2. This ranking helps users find the best matches quickly.

Practice

1. What does relevance scoring in Elasticsearch primarily determine?

easy

A. How well a document matches the search query

B. The size of the document

C. The date when the document was created

D. The number of fields in the document

Solution

1. Step 1: Understand relevance scoring purpose

   Relevance scoring measures how closely a document matches the search query terms.

2. Step 2: Compare options to scoring concept

   Only How well a document matches the search query relates to matching quality, others are unrelated to scoring.

3. Final Answer:

   How well a document matches the search query -> Option A

4. Quick Check:

   Relevance scoring = match quality [OK]

Hint: Relevance scoring ranks by match quality, not metadata [OK]

Common Mistakes:

• Confusing scoring with document size
• Thinking scoring depends on creation date
• Assuming number of fields affects score

2. Which Elasticsearch query part directly influences relevance scoring?

easy

A. sort

B. match

C. filter

D. aggs

Solution

1. Step 1: Identify query parts affecting scoring

   The match query scores documents based on how well they match the query text.

2. Step 2: Differentiate from other parts

   filter and aggs do not affect scoring; sort orders results but doesn't calculate score.

3. Final Answer:

   match -> Option B

4. Quick Check:

   Match query = scoring influence [OK]

Hint: Match queries affect scoring; filters do not [OK]

Common Mistakes:

• Thinking filters affect scoring
• Confusing sort with scoring
• Assuming aggregations influence score

3. Given this Elasticsearch query snippet, which document will rank higher?

{
  "query": {
    "match": {
      "title": "apple"
    }
  }
}

Documents: 1. {"title": "apple pie recipe"} 2. {"title": "banana apple smoothie"} 3. {"title": "fruit salad"}

medium

A. Document 3

B. Document 2

C. Document 1

D. All have the same score

Solution

1. Step 1: Understand match query scoring

   Documents containing the exact term "apple" in the title get higher scores. Document 1 has "apple" at the start and is more focused.

2. Step 2: Compare documents by term presence

   Document 1 has "apple" prominently; Document 2 has "apple" but mixed with "banana"; Document 3 lacks "apple".

3. Final Answer:

   Document 1 -> Option C

4. Quick Check:

   Exact term presence = higher score [OK]

Hint: Exact term in title ranks higher than partial or missing [OK]

Common Mistakes:

• Assuming all documents score equally
• Choosing document without the term
• Ignoring term position and frequency

4. You wrote this Elasticsearch query but results are not ranked by relevance:

{
  "query": {
    "bool": {
      "filter": {
        "match": { "content": "coffee" }
      }
    }
  }
}

What is the main issue?

medium

A. "filter" does not calculate relevance scores

B. The "match" query is misspelled

C. The query must use "must_not" instead of "filter"

D. The field "content" does not exist

Solution

1. Step 1: Understand bool query filter behavior

   The filter clause filters documents but does not affect relevance scoring.

2. Step 2: Identify why scoring is missing

   Using filter with match disables scoring; to rank by relevance, use must instead.

3. Final Answer:

   "filter" does not calculate relevance scores -> Option A

4. Quick Check:

   Filter disables scoring, use must for relevance [OK]

Hint: Use must, not filter, for scoring queries [OK]

Common Mistakes:

• Thinking filter affects scoring
• Assuming match is misspelled
• Ignoring bool query clause differences

5. You want to rank documents by relevance but also exclude those with "draft" status. Which query correctly ranks by relevance and filters out drafts?

{
  "query": {
    "bool": {
      "must": {
        "match": { "text": "Elasticsearch" }
      },
      "filter": {
        "term": { "status": "draft" }
      }
    }
  }
}

hard

A. Use "filter" for both match and filtering drafts

B. Use "filter" for match and "must" for filtering drafts

C. Use "should" for both match and filtering drafts

D. Use "must" for match and "must_not" for filtering drafts

Solution

1. Step 1: Understand must and filter roles

   must runs scoring queries; filter filters without scoring; must_not excludes documents.

2. Step 2: Apply exclusion for drafts

   To exclude drafts, use must_not with a term filter on status "draft"; keep must for scoring match query.

3. Final Answer:

   Use "must" for match and "must_not" for filtering drafts -> Option D

4. Quick Check:

   Must scores, must_not excludes [OK]

Hint: Use must for scoring, must_not to exclude unwanted docs [OK]

Common Mistakes:

• Using filter to exclude documents (doesn't exclude)
• Using should which affects scoring differently
• Filtering match query disables scoring