Elasticsearch

Elasticsearch 是一个分布式的、基于 REST 的搜索和分析引擎。它提供了一个分布式、多租户能力的全文搜索引擎，具有 HTTP Web 接口和无模式的 JSON 文档。它支持关键词搜索、向量搜索、混合搜索和复杂过滤。

ElasticsearchRetriever 是一个通用的包装器，通过Query DSL实现对所有Elasticsearch功能的灵活访问。对于大多数用例，其他类（如ElasticsearchStore、ElasticsearchEmbeddings等）应该足够了，但如果它们不够用，你可以使用ElasticsearchRetriever。

%pip install --upgrade --quiet elasticsearch langchain-elasticsearch

from typing import Any, Dict, Iterable
from elasticsearch import Elasticsearch
from elasticsearch.helpers import bulk
from langchain_community.embeddings import DeterministicFakeEmbedding
from langchain_core.documents import Document
from langchain_core.embeddings import Embeddings
from langchain_elasticsearch import ElasticsearchRetriever

配置

在这里，我们定义了与 Elasticsearch 的连接。在这个示例中，我们使用本地运行的实例。或者，你可以在Elastic Cloud上创建一个帐户并开始免费试用。

es_url = "http://localhost:9200"
es_client = Elasticsearch(hosts=[es_url])
es_client.info()

对于向量搜索，我们将使用随机嵌入，仅用于说明。对于实际用例，请选择可用的 LangChain Embeddings 类之一。

embeddings = DeterministicFakeEmbedding(size=3)

定义示例数据

index_name = "test-langchain-retriever"
text_field = "text"
dense_vector_field = "fake_embedding"
num_characters_field = "num_characters"
texts = [
    "foo",
    "bar",
    "world",
    "hello world",
    "hello",
    "foo bar",
    "bla bla foo",
]

索引数据

通常，用户在 Elasticsearch 索引中已经有数据时使用 ElasticsearchRetriever。在这里，我们索引了一些示例文本文档。如果你已经使用 ElasticsearchStore.from_documents 创建了一个索引，那也可以。

def create_index(
    es_client: Elasticsearch,
    index_name: str,
    text_field: str,
    dense_vector_field: str,
    num_characters_field: str,
):
    es_client.indices.create(
        index=index_name,
        mappings={
            "properties": {
                text_field: {"type": "text"},
                dense_vector_field: {"type": "dense_vector"},
                num_characters_field: {"type": "integer"},
            }
        },
    )
def index_data(
    es_client: Elasticsearch,
    index_name: str,
    text_field: str,
    dense_vector_field: str,
    embeddings: Embeddings,
    texts: Iterable[str],
    refresh: bool = True,
) -> None:
    create_index(
        es_client, index_name, text_field, dense_vector_field, num_characters_field
    )
    vectors = embeddings.embed_documents(list(texts))
    requests = [
        {
            "_op_type": "index",
            "_index": index_name,
            "_id": i,
            text_field: text,
            dense_vector_field: vector,
            num_characters_field: len(text),
        }
        for i, (text, vector) in enumerate(zip(texts, vectors))
    ]
    bulk(es_client, requests)
    if refresh:
        es_client.indices.refresh(index=index_name)
    return len(requests)

index_data(es_client, index_name, text_field, dense_vector_field, embeddings, texts)

7

使用示例

向量搜索

在这个示例中，使用虚构的嵌入进行密集向量检索。

def vector_query(search_query: str) -> Dict:
    vector = embeddings.embed_query(search_query)  # 与索引相同的嵌入
    return {
        "knn": {
            "field": dense_vector_field,
            "query_vector": vector,
            "k": 5,
            "num_candidates": 10,
        }
    }
vector_retriever = ElasticsearchRetriever.from_es_params(
    index_name=index_name,
    body_func=vector_query,
    content_field=text_field,
    url=es_url,
)
vector_retriever.invoke("foo")

[Document(page_content='foo', metadata={'_index': 'test-langchain-index', '_id': '0', '_score': 1.0, '_source': {'fake_embedding': [-2.336764233933763, 0.27510289545940503, -0.7957597268194339], 'num_characters': 3}}),
 Document(page_content='world', metadata={'_index': 'test-langchain-index', '_id': '2', '_score': 0.6770179, '_source': {'fake_embedding': [-0.7041151202179595, -1.4652961969276497, -0.25786766898672847], 'num_characters': 5}}),
 Document(page_content='hello world', metadata={'_index': 'test-langchain-index', '_id': '3', '_score': 0.4816144, '_source': {'fake_embedding': [0.42728413221815387, -1.1889908285425348, -1.445433230084671], 'num_characters': 11}}),
 Document(page_content='hello', metadata={'_index': 'test-langchain-index', '_id': '4', '_score': 0.46853775, '_source': {'fake_embedding': [-0.28560441330564046, 0.9958894823084921, 1.5489829880195058], 'num_characters': 5}}),
 Document(page_content='foo bar', metadata={'_index': 'test-langchain-index', '_id': '5', '_score': 0.2086992, '_source': {'fake_embedding': [0.2533670476638539, 0.08100381646160418, 0.7763644080870179], 'num_characters': 7}})]

BM25

传统的关键词匹配。

def bm25_query(search_query: str) -> Dict:
    return {
        "query": {
            "match": {
                text_field: search_query,
            },
        },
    }
bm25_retriever = ElasticsearchRetriever.from_es_params(
    index_name=index_name,
    body_func=bm25_query,
    content_field=text_field,
    url=es_url,
)
bm25_retriever.invoke("foo")

[Document(page_content='foo', metadata={'_index': 'test-langchain-index', '_id': '0', '_score': 0.9711467, '_source': {'fake_embedding': [-2.336764233933763, 0.27510289545940503, -0.7957597268194339], 'num_characters': 3}}),
 Document(page_content='foo bar', metadata={'_index': 'test-langchain-index', '_id': '5', '_score': 0.7437035, '_source': {'fake_embedding': [0.2533670476638539, 0.08100381646160418, 0.7763644080870179], 'num_characters': 7}}),
 Document(page_content='bla bla foo', metadata={'_index': 'test-langchain-index', '_id': '6', '_score': 0.6025789, '_source': {'fake_embedding': [1.7365927060137358, -0.5230400847844948, 0.7978339724186192], 'num_characters': 11}})]

混合搜索

使用Reciprocal Rank Fusion (RRF) 结合向量搜索和BM25搜索的结果集。

def hybrid_query(search_query: str) -> Dict:
    vector = embeddings.embed_query(search_query)  # 与索引相同的嵌入
    return {
        "query": {
            "match": {
                text_field: search_query,
            },
        },
        "knn": {
            "field": dense_vector_field,
            "query_vector": vector,
            "k": 5,
            "num_candidates": 10,
        },
        "rank": {"rrf": {}},
    }
hybrid_retriever = ElasticsearchRetriever.from_es_params(
    index_name=index_name,
    body_func=hybrid_query,
    content_field=text_field,
    url=es_url,
)
hybrid_retriever.invoke("foo")

[Document(page_content='foo', metadata={'_index': 'test-langchain-index', '_id': '0', '_score': 0.9711467, '_source': {'fake_embedding': [-2.336764233933763, 0.27510289545940503, -0.7957597268194339], 'num_characters': 3}}),
 Document(page_content='foo bar', metadata={'_index': 'test-langchain-index', '_id': '5', '_score': 0.7437035, '_source': {'fake_embedding': [0.2533670476638539, 0.08100381646160418, 0.7763644080870179], 'num_characters': 7}}),
 Document(page_content='bla bla foo', metadata={'_index': 'test-langchain-index', '_id': '6', '_score': 0.6025789, '_source': {'fake_embedding': [1.7365927060137358, -0.5230400847844948, 0.7978339724186192], 'num_characters': 11}})]

模糊匹配

具有容错能力的关键词匹配。

def fuzzy_query(search_query: str) -> Dict:
    return {
        "query": {
            "match": {
                text_field: {
                    "query": search_query,
                    "fuzziness": "AUTO",
                }
            },
        },
    }
fuzzy_retriever = ElasticsearchRetriever.from_es_params(
    index_name=index_name,
    body_func=fuzzy_query,
    content_field=text_field,
    url=es_url,
)
fuzzy_retriever.invoke("fox")  # 注意字符容差

[Document(page_content='foo', metadata={'_index': 'test-langchain-index', '_id': '0', '_score': 0.6474311, '_source': {'fake_embedding': [-2.336764233933763, 0.27510289545940503, -0.7957597268194339], 'num_characters': 3}}),
 Document(page_content='foo bar', metadata={'_index': 'test-langchain-index', '_id': '5', '_score': 0.49580228, '_source': {'fake_embedding': [0.2533670476638539, 0.08100381646160418, 0.7763644080870179], 'num_characters': 7}}),
 Document(page_content='bla bla foo', metadata={'_index': 'test-langchain-index', '_id': '6', '_score': 0.40171927, '_source': {'fake_embedding': [1.7365927060137358, -0.5230400847844948, 0.7978339724186192], 'num_characters': 11}})]

复杂过滤

在不同字段上组合过滤器。

def filter_query_func(search_query: str) -> Dict:
    return {
        "query": {
            "bool": {
                "must": [
                    {"range": {num_characters_field: {"gte": 5}}},
                ],
                "must_not": [
                    {"prefix": {text_field: "bla"}},
                ],
                "should": [
                    {"match": {text_field: search_query}},
                ],
            }
        }
    }
filtering_retriever = ElasticsearchRetriever.from_es_params(
    index_name=index_name,
    body_func=filter_query_func,
    content_field=text_field,
    url=es_url,
)
filtering_retriever.invoke("foo")

output

自定义文档映射器

可以自定义将 Elasticsearch 结果（命中）映射为 LangChain 文档的函数。

def num_characters_mapper(hit: Dict[str, Any]) -> Document:
    num_chars = hit["_source"][num_characters_field]
    content = hit["_source"][text_field]
    return Document(
        page_content=f"This document has {num_chars} characters",
        metadata={"text_content": content},
    )
custom_mapped_retriever = ElasticsearchRetriever.from_es_params(
    index_name=index_name,
    body_func=filter_query_func,
    document_mapper=num_characters_mapper,
    url=es_url,
)
custom_mapped_retriever.invoke("foo")

[Document(page_content='This document has 7 characters', metadata={'text_content': 'foo bar'}),
 Document(page_content='This document has 5 characters', metadata={'text_content': 'world'}),
 Document(page_content='This document has 11 characters', metadata={'text_content': 'hello world'}),
 Document(page_content='This document has 5 characters', metadata={'text_content': 'hello'})]

注意，查询匹配结果在顶部。通过过滤器的其他文档也在结果集中，但它们的得分都相同。