Documentation
¶
Overview ¶
Package reranker provides document re-ranking functionality for improving search quality.
Index ¶
Constants ¶
This section is empty.
Variables ¶
var ErrNilContext = errors.New("context cannot be nil")
ErrNilContext is returned when a nil context is passed to Rerank.
Functions ¶
This section is empty.
Types ¶
type Document ¶
type Document struct {
ID string // Unique identifier for the document
Content string // Text content to be re-ranked
Score float32 // Original similarity score from search
}
Document represents a searchable document with metadata and scores.
type Reranker ¶
type Reranker interface {
// Rerank re-ranks documents based on query relevance.
// Takes a query string, list of documents, and desired top K results.
// Returns re-ranked documents sorted by RerankerScore in descending order,
// limited to topK results.
//
// The caller is responsible for ensuring ctx is not nil.
Rerank(ctx context.Context, query string, docs []Document, topK int) ([]ScoredDocument, error)
// Close closes the reranker and releases any resources.
// Should be called when the reranker is no longer needed.
Close() error
}
Reranker provides an interface for document re-ranking algorithms.
type ScoredDocument ¶
type ScoredDocument struct {
Document
RerankerScore float32 // Score from re-ranker (0.0-1.0)
OriginalRank int // Original rank position in results (0-indexed)
}
ScoredDocument represents a document with re-ranking scores.
type SimpleReranker ¶
type SimpleReranker struct{}
SimpleReranker implements a simple TF-IDF based reranking algorithm. It calculates term overlap between the query and documents, then combines the original score with the overlap score to produce a final ranking.
func NewSimpleReranker ¶
func NewSimpleReranker() *SimpleReranker
NewSimpleReranker creates a new SimpleReranker instance.
func (*SimpleReranker) Close ¶
func (r *SimpleReranker) Close() error
Close closes the reranker. SimpleReranker has no resources to clean up.
func (*SimpleReranker) Rerank ¶
func (r *SimpleReranker) Rerank(ctx context.Context, query string, docs []Document, topK int) ([]ScoredDocument, error)
Rerank re-ranks documents using a simple TF-IDF approach. The algorithm: 1. Tokenizes the query into lowercased terms 2. For each document, calculates term overlap with the query 3. Combines original score (50% weight) with overlap score (50% weight) 4. Sorts by combined score and returns top K results
Source Files