feat(): add elasticsearch hybrid search

libs/langchain-community/src/vectorstores/elasticsearch.ts

@@ -3,6 +3,7 @@ import { Client, estypes } from "@elastic/elasticsearch";
 import type { EmbeddingsInterface } from "@langchain/core/embeddings";
 import { VectorStore } from "@langchain/core/vectorstores";
 import { Document } from "@langchain/core/documents";
+import type { Callbacks } from "@langchain/core/callbacks/manager";
 /**
  * Type representing the k-nearest neighbors (k-NN) engine used in
  * Elasticsearch.
@@ -24,6 +25,36 @@ interface VectorSearchOptions {
   readonly candidates?: number;
 }
 
+/**
+ * Configuration options for hybrid retrieval strategy.
+ */
+export interface HybridRetrievalStrategyConfig {
+  rankWindowSize?: number;
+  rankConstant?: number;
+  textField?: string;
+  /**
+   * For Elasticsearch 9.x, set to `false` to include vectors in responses.
+   */
+  excludeSourceVectors?: boolean;
+}
+
+/**
+ * Hybrid search strategy combining vector and BM25 search using RRF.
+ */
+export class HybridRetrievalStrategy {
+  public readonly rankWindowSize: number;
+  public readonly rankConstant: number;
+  public readonly textField: string;
+  public readonly excludeSourceVectors?: boolean;
+
+  constructor(config: HybridRetrievalStrategyConfig = {}) {
+    this.rankWindowSize = config.rankWindowSize ?? 100;
+    this.rankConstant = config.rankConstant ?? 60;
+    this.textField = config.textField ?? "text";
+    this.excludeSourceVectors = config.excludeSourceVectors;
+  }
+}
+
 /**
  * Interface defining the arguments required to create an Elasticsearch
  * client.
@@ -32,6 +63,7 @@ export interface ElasticClientArgs {
   readonly client: Client;
   readonly indexName?: string;
   readonly vectorSearchOptions?: VectorSearchOptions;
+  readonly strategy?: HybridRetrievalStrategy;
 }
 
 /**
@@ -73,6 +105,10 @@ export class ElasticVectorSearch extends VectorStore {
 
   private readonly candidates: number;
 
+  private readonly strategy?: HybridRetrievalStrategy;
+
+  private lastQueryText?: string;
+
   _vectorstoreType(): string {
     return "elasticsearch";
   }
@@ -85,9 +121,14 @@ export class ElasticVectorSearch extends VectorStore {
     this.m = args.vectorSearchOptions?.m ?? 16;
     this.efConstruction = args.vectorSearchOptions?.efConstruction ?? 100;
     this.candidates = args.vectorSearchOptions?.candidates ?? 200;
+    this.strategy = args.strategy;
+
+    const userAgent = this.strategy
+      ? "langchain-js-vs-hybrid/0.0.1"
+      : "langchain-js-vs/0.0.1";
 
     this.client = args.client.child({
-      headers: { "user-agent": "langchain-js-vs/0.0.1" },
+      headers: { "user-agent": userAgent },
     });
     this.indexName = args.indexName ?? "documents";
   }
@@ -155,6 +196,16 @@ export class ElasticVectorSearch extends VectorStore {
     return documentIds;
   }
 
+  async similaritySearch(
+    query: string,
+    k = 4,
+    filter?: ElasticFilter,
+    _callbacks?: Callbacks
+  ): Promise<Document[]> {
+    this.lastQueryText = query;
+    return super.similaritySearch(query, k, filter, _callbacks);
+  }
+
   /**
    * Method to perform a similarity search in the Elasticsearch database
    * using a vector. It returns the k most similar documents along with
@@ -315,6 +366,12 @@ export class ElasticVectorSearch extends VectorStore {
       },
     };
 
+    if (this.strategy?.excludeSourceVectors !== undefined) {
+      request.settings = {
+        "index.mapping.exclude_source_vectors": this.strategy.excludeSourceVectors,
+      };
+    }
+
     const indexExists = await this.doesIndexExist();
     if (indexExists) return;