BM25

Defined in: packages/core/src/vector-store/bm25.ts:24

A simple BM25 (Best Matching 25) implementation for in-memory search.

BM25 is a bag-of-words retrieval function that ranks documents based on the query terms appearing in each document. It’s an improvement over TF-IDF that includes document length normalization.

Key parameters:

• k1 (default 1.5): Controls term frequency saturation. Higher values give more weight to term frequency.
• b (default 0.75): Controls document length normalization. 0 = no normalization, 1 = full normalization.

Example

const bm25 = new BM25(documents);
const results = bm25.search("search query", 10);
// Returns top 10 documents with their BM25 scores

Constructors

Constructor

new BM25(nodes, options?): BM25

Defined in: packages/core/src/vector-store/bm25.ts:41

Creates a new BM25 index from the given nodes.

Parameters

nodes

BaseNode<Metadata>[]

Array of nodes to index

options?

Optional BM25 parameters

k1?

number

Term frequency saturation parameter (default: 1.5)

b?

number

Document length normalization parameter (default: 0.75)

Returns

BM25

Methods

search()

search(query, topK): object[]

Defined in: packages/core/src/vector-store/bm25.ts:96

Searches the index for documents matching the query.

Parameters

query

string

The search query string

topK

number

Maximum number of results to return

Returns

object[]

Array of document IDs with their BM25 scores, sorted by score descending

---

getDocumentCount()

getDocumentCount(): number

Defined in: packages/core/src/vector-store/bm25.ts:130

Returns the number of documents in the index.

Returns

number

---

getAverageDocumentLength()

getAverageDocumentLength(): number

Defined in: packages/core/src/vector-store/bm25.ts:137

Returns the average document length in the index.

Returns

number