hybridSearch - Hybrid search

hybridSearch() combines full-text search and vector similarity search with ranking.

info

This API can only be used when connected with SeekdbClient. For more information about SeekdbClient, see SeekdbClient.

Prerequisites

• You have installed seekdb-js. For more information, see Quick start.

• You have installed the server mode of seekdb. For more information, see Deploy seekdb by using yum install.

• You are connected to the database. For more information, see SeekdbClient.

• You have created a collection and inserted corresponding data. For more information, see createCollection and add.

Request parameters

hybrid_search(
    query={
        "whereDocument": ,
        "where": ,
        "nResults": 
    },
    knn={
        queryEmbeddings
        "queryTexts": 
        "where":
        "nResults": 
    },
    rank=,  
    nResults=,
    include=
)

• query: The full-text search configuration, which contains the following parameters:

  Parameter	Type	Required	Description	Example value
  where	Where	No	The metadata filter condition.	{"category": {"$eq": "AI"}}
  whereDocument	WhereDocument	No	The document content filter condition (full-text search).	{"$contains": "machine"}
  nResults	number	No	The number of results returned by full-text search.

• knn: The vector search configuration, which contains the following parameters:

  Parameter	Type	Required	Description	Example value
  queryEmbeddings	number[] | number[][]	No	The single vector or vector list for batch queries. If provided, it will be used directly (ignoring embeddingFunction). If not provided, query_text must be provided, and collection must have embeddingFunction	[1.0, 2.0, 3.0]
  queryTexts	string | string[]	No	The single vector0 or vector list. If provided, it will be used directly (ignoring embeddingFunction). If not provided, documents must be provided, and collection must have embeddingFunction.	["my query text"]
  where	Where	No	The metadata filter condition.	{"category": {"$eq": "AI"}}
  nResults	number	No	The number of results returned by vector search.

• Other parameters are as follows:

  |Parameter|Type|Required|Description|Example value| |rank|HybridSearchRank |No|The ranking configuration. For example: {"rrf": {"rank_window_size": 60, "rank_constant": 60}}|{"category": {"$eq": "AI"}}| |nResults|number|Yes|The number of similar results to return. Default value: 10|3| |include|readonly ("documents" | "metadatas" | "embeddings" | "distances")[]|No|The list of fields to include: ["documents", "metadatas", "embeddings"].|["documents", "metadatas", "embeddings"]|

info

• The embeddingFunction used is associated with the collection (set during createCollection() or getCollection()). You cannot override it for each operation.
• At least one of query or knn must be provided.
• knn:
  • If queryEmbeddings is provided, the dimensions will be validated against collection.dimension.
  • queryTexts: string or string[]; the embeddingFunction of the collection is used to automatically embed (if the function is missing, a ValueError is returned).
• If knn.queryTexts is provided, embeddingFunction must be configured.

Request example

import { SeekdbClient } from "seekdb";

const client = new SeekdbClient({
  host: "127.0.0.1",
  port: 2881,
  user: "root",
  password: "",
  database: "test",
});

const collection = await client.getCollection({name: 'my_collection'});

const results = await  collection.hybridSearch({
    query: {
        whereDocument: {$contains: "machine learning"},
        nResults: 10
    },
        knn: {
        queryTexts: 'documents',
        nResults: 10
    },
    rank: {rrf: {}},
    nResults: 10
});

Response parameters

The search result dictionary, which contains the ID, distances, metadatas, document, and embeddings.

Related operations

• Vector query
• Retrieval
• Operators