Skip to main content

​
Field Boosting

Boosting allows you to give more weight to specific properties during search, influencing the relevance score of documents. This is useful when certain fields are more important than others for your search use case.

​
Basic Boosting

Use the boost parameter to multiply the score of specific properties: 
import { create, insert, search } from '@orama/orama'

const db = create({
  schema: {
    title: 'string',
    author: 'string',
    description: 'string'
  }
})

insert(db, {
  title: 'Introduction to Orama',
  author: 'Michael',
  description: 'Learn about Orama search engine'
})

const results = search(db, {
  term: 'Michael',
  properties: ['title', 'author', 'description'],
  boost: {
    title: 2  // Title matches are 2x more important
  }
})

​
How Boosting Works

When you boost a property, Orama multiplies the BM25 score for that property by the boost value:
1

Calculate Base Score

Orama calculates the BM25 score for each property where the term appears
2

Apply Boost Multiplier

The score for boosted properties is multiplied by the boost value
3

Combine Scores

All property scores are combined to create the final document score
4

Rank Results

Documents are ranked by their final combined score

​
Boost Values

Boost values are numeric multipliers: 
const results = search(db, {
  term: 'headphones',
  boost: {
    title: 2  // Title matches count twice as much
  }
})
  • Values greater than 1 increase importance
  • Values between 0 and 1 decrease importance
  • Value of 1 has no effect (default)
  • Negative values are not allowed

​
Multiple Field Boosting

Boost multiple properties with different values: 
const results = search(db, {
  term: 'JavaScript tutorials',
  properties: ['title', 'description', 'tags', 'author'],
  boost: {
    title: 3,        // Titles are most important
    tags: 2,         // Tags are moderately important
    description: 1,  // Description has normal weight
    author: 0.5      // Author is least important
  }
})

​
Nested Property Boosting

Boost nested object properties: 
const db = create({
  schema: {
    title: 'string',
    content: {
      heading: 'string',
      body: 'string',
      footer: 'string'
    }
  }
})

const results = search(db, {
  term: 'search',
  boost: {
    title: 3,
    'content.heading': 2,
    'content.body': 1,
    'content.footer': 0.5
  }
})

​
Practical Examples

Prioritize product names over descriptions: 
const db = create({
  schema: {
    name: 'string',
    brand: 'string',
    description: 'string',
    category: 'string'
  }
})

const results = search(db, {
  term: 'wireless headphones',
  boost: {
    name: 3,         // Product name is most important
    brand: 2,        // Brand is important
    category: 1.5,   // Category is moderately important
    description: 1   // Description has normal weight
  }
})
Prioritize titles and headings: 
const db = create({
  schema: {
    title: 'string',
    subtitle: 'string',
    content: 'string',
    tags: 'string[]'
  }
})

const results = search(db, {
  term: 'authentication',
  boost: {
    title: 5,      // Titles are highly important
    subtitle: 3,   // Subtitles are important
    tags: 2,       // Tags are moderately important
    content: 1     // Content has normal weight
  }
})

​
Blog Post Search

Balance title, author, and content: 
const db = create({
  schema: {
    title: 'string',
    author: 'string',
    excerpt: 'string',
    body: 'string',
    tags: 'string[]'
  }
})

const results = search(db, {
  term: 'machine learning',
  boost: {
    title: 4,     // Titles most important
    tags: 2.5,    // Tags very important
    excerpt: 1.5, // Excerpts moderately important
    author: 1,    // Author normal weight
    body: 0.8     // Body slightly less important
  }
})

​
Combining Boosting with Other Features

​
With Property Filtering

const results = search(db, {
  term: 'headphones',
  properties: ['title', 'description'],  // Only search these fields
  boost: {
    title: 2  // Boost only applies to searched properties
  }
})
Boost values are only applied to properties included in the properties array. Boosting a property not in the search will have no effect.

​
With Filters

const results = search(db, {
  term: 'laptop',
  boost: {
    title: 3,
    brand: 2
  },
  where: {
    price: {
      lt: 1000
    },
    inStock: true
  }
})

​
With BM25 Tuning

const results = search(db, {
  term: 'headphones',
  boost: {
    title: 2
  },
  relevance: {
    k: 1.2,   // BM25 term frequency saturation
    b: 0.75,  // BM25 document length normalization
    d: 0.5    // BM25 frequency normalization
  }
})

​
With Exact Matching

const results = search(db, {
  term: 'Noise Cancelling Headphones',
  exact: true,  // Exact phrase matching
  boost: {
    title: 3  // Boost still applies to exact matches
  }
})
Boosting only affects the full-text search component of hybrid search: 
const results = search(db, {
  mode: 'hybrid',
  term: 'headphones',
  vector: {
    value: embeddings,
    property: 'embedding'
  },
  boost: {
    title: 2  // Only affects full-text search, not vector search
  },
  hybridWeights: {
    text: 0.7,    // Weight for full-text results (with boosting)
    vector: 0.3   // Weight for vector results (no boosting)
  }
})
Boost parameters only apply to the full-text search portion of hybrid search. Vector search results are not affected by field boosting.

​
Dynamic Boosting

Adjust boost values based on user context or preferences: 
function searchProducts(term, userPreferences) {
  const boost = {
    name: 2,
    description: 1
  }
  
  // Increase brand boost if user is brand-conscious
  if (userPreferences.preferBrands) {
    boost.brand = 3
  }
  
  // Increase category boost if searching within category
  if (userPreferences.categoryFocus) {
    boost.category = 2.5
  }
  
  return search(db, {
    term,
    boost
  })
}

​
Testing Boost Values

Experiment with different boost values to find the optimal configuration: 
// Test different configurations
const configs = [
  { title: 2, description: 1 },
  { title: 3, description: 1 },
  { title: 5, description: 1 },
  { title: 3, description: 0.5 }
]

configs.forEach((boost, i) => {
  const results = search(db, {
    term: 'test query',
    boost,
    limit: 10
  })
  
  console.log(`Config ${i + 1}:`, results.hits[0]?.score)
})

​
Performance Considerations

1

Boosting is Efficient

Boosting is a simple multiplication operation and has minimal performance impact
2

Limit Boosted Properties

While you can boost many properties, focus on the most important ones:
// Good - focused boosting
boost: { title: 3, subtitle: 2 }

// Less optimal - many boosted properties
boost: { field1: 2, field2: 1.5, field3: 1.3, field4: 1.1 }
3

Combine with Property Filtering

Search only necessary properties to improve performance:
search(db, {
  term: 'query',
  properties: ['title', 'description'],  // Limit search scope
  boost: { title: 2 }
})

​
Best Practices

Start with 2-3x

Begin with moderate boost values (2-3x) and adjust based on results

Use Consistently

Apply similar boosting strategies across similar search types

Test with Real Queries

Validate boost values with actual user search queries

Document Choices

Document why certain fields are boosted for future reference

Full-Text Search

Learn about full-text search fundamentals

Hybrid Search

Combine boosting with hybrid search

Sorting

Sort boosted search results

Filters

Combine boosting with filters

Build docs developers (and LLMs) love

Get started for freeTalk to us