Title-Scorer

Title scorer is a simple scorer for the "Default" role that matches documents by applying Levenshtein distance to document titles. This is the default scoring method for basic document ranking.

Available Scoring Methods

Similarity-Based Scoring (Default)

• Levenshtein Distance: Measures edit distance between query and document title
• Jaro Distance: String similarity based on character transpositions
• Jaro-Winkler Distance: Enhanced Jaro distance with prefix bonus

BM25-Based Scoring (New)

Terraphim now supports multiple BM25 variants for more sophisticated document ranking:

• BM25: Standard Okapi BM25 ranking function
• BM25F: Fielded BM25 with different weights for title, body, description, tags
• BM25Plus: Enhanced BM25 with additional parameters for fine-tuning
• TFIDF: Traditional Term Frequency-Inverse Document Frequency
• Jaccard: Similarity-based scoring using term overlap
• QueryRatio: Ratio of query terms matched in document

Configuration

You can select different scoring methods via the QueryScorer enum:

{
  "relevance_function": "title-scorer",
  "query_scorer": "bm25f"
}

Usage

The title scorer is automatically used when RelevanceFunction::TitleScorer is selected in your role configuration. The system will:

1. Index all documents using the selected scoring method
2. Apply the chosen algorithm to rank search results
3. Return documents sorted by relevance score

When to Use

• Similarity-based scoring: For simple title matching and fuzzy search
• BM25 variants: For more sophisticated document ranking that considers term frequency and document structure
• BM25F: When you want to emphasize certain document fields (title, body, description, tags)
• TFIDF: For simple, interpretable scoring without length bias
• Jaccard/QueryRatio: When you care about term presence rather than frequency

For detailed information about BM25 scorers, see BM25 Scorers.