Skip to main content
Neo4j 是由 Neo4j, Inc 开发的图数据库管理系统。
Neo4j 存储的数据元素包括节点、连接节点的边以及节点和边的属性。其开发者将其描述为具有原生图存储和处理功能的符合 ACID 的事务型数据库。Neo4j 提供非开源的“社区版”，采用修改版的 GNU 通用公共许可证授权，在线备份和高可用性扩展则采用闭源商业许可证授权。Neo 也以闭源商业条款授权包含这些扩展的 Neo4j。
本笔记本演示如何使用大语言模型（LLM）为可通过 Cypher 查询语言进行查询的图数据库提供自然语言接口。
Cypher 是一种声明式图查询语言，允许在属性图中进行表达力强且高效的数据查询。

​
设置环境

你需要一个正在运行的 Neo4j 实例。一种选择是在其 Aura 云服务中创建一个免费的 Neo4j 数据库实例。你也可以使用 Neo4j Desktop 应用程序 在本地运行该数据库，或者运行 Docker 容器。 你可以通过执行以下脚本来运行本地 Docker 容器： 
docker run \
    --name neo4j \
    -p 7474:7474 -p 7687:7687 \
    -d \
    -e NEO4J_AUTH=neo4j/password \
    -e NEO4J_PLUGINS=\[\"apoc\"\]  \
    neo4j:latest
如果你使用的是 Docker 容器，需要等待几秒钟让数据库启动。 
from langchain_neo4j import GraphCypherQAChain, Neo4jGraph
from langchain_openai import ChatOpenAI

graph = Neo4jGraph(url="bolt://localhost:7687", username="neo4j", password="password")
本指南默认使用 OpenAI 模型。 
import getpass
import os

if "OPENAI_API_KEY" not in os.environ:
    os.environ["OPENAI_API_KEY"] = getpass.getpass("OpenAI API Key:")

​
填充数据库

假设你的数据库为空，你可以使用 Cypher 查询语言对其进行填充。以下 Cypher 语句具有幂等性，这意味着无论执行一次还是多次，数据库信息都将保持一致。 
graph.query(
    """
MERGE (m:Movie {name:"Top Gun", runtime: 120})
WITH m
UNWIND ["Tom Cruise", "Val Kilmer", "Anthony Edwards", "Meg Ryan"] AS actor
MERGE (a:Actor {name:actor})
MERGE (a)-[:ACTED_IN]->(m)
"""
)

[]

​
刷新图模式信息

如果数据库模式发生变化，你可以刷新生成 Cypher 语句所需的模式信息。 
graph.refresh_schema()

print(graph.schema)

Node properties:
Movie {runtime: INTEGER, name: STRING}
Actor {name: STRING}
Relationship properties:

The relationships:
(:Actor)-[:ACTED_IN]->(:Movie)

​
增强模式信息

选择增强模式版本可使系统自动扫描数据库中的示例值并计算一些分布指标。例如，如果某个节点属性的不同值少于 10 个，我们将在模式中返回所有可能的值。否则，每个节点和关系属性仅返回单个示例值。 
enhanced_graph = Neo4jGraph(
    url="bolt://localhost:7687",
    username="neo4j",
    password="password",
    enhanced_schema=True,
)
print(enhanced_graph.schema)

Node properties:
- **Movie**
  - `runtime`: INTEGER Min: 120, Max: 120
  - `name`: STRING Available options: ['Top Gun']
- **Actor**
  - `name`: STRING Available options: ['Tom Cruise', 'Val Kilmer', 'Anthony Edwards', 'Meg Ryan']
Relationship properties:

The relationships:
(:Actor)-[:ACTED_IN]->(:Movie)

​
查询图数据

我们现在可以使用图 Cypher QA 链向图提问。 
chain = GraphCypherQAChain.from_llm(
    ChatOpenAI(temperature=0), graph=graph, verbose=True, allow_dangerous_requests=True
)

chain.invoke({"query": "Who played in Top Gun?"})

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE m.name = 'Top Gun'
RETURN a.name
Full Context:
[{'a.name': 'Tom Cruise'}, {'a.name': 'Val Kilmer'}, {'a.name': 'Anthony Edwards'}, {'a.name': 'Meg Ryan'}]

> Finished chain.

{'query': 'Who played in Top Gun?',
 'result': 'Tom Cruise, Val Kilmer, Anthony Edwards, and Meg Ryan played in Top Gun.'}

​
限制结果数量

你可以使用 top_k 参数限制 Cypher QA 链返回的结果数量。 默认值为 10。 
chain = GraphCypherQAChain.from_llm(
    ChatOpenAI(temperature=0),
    graph=graph,
    verbose=True,
    top_k=2,
    allow_dangerous_requests=True,
)

chain.invoke({"query": "Who played in Top Gun?"})

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE m.name = 'Top Gun'
RETURN a.name
Full Context:
[{'a.name': 'Tom Cruise'}, {'a.name': 'Val Kilmer'}]

> Finished chain.

{'query': 'Who played in Top Gun?',
 'result': 'Tom Cruise, Val Kilmer played in Top Gun.'}

​
返回中间结果

你可以使用 return_intermediate_steps 参数返回 Cypher QA 链的中间步骤。 
chain = GraphCypherQAChain.from_llm(
    ChatOpenAI(temperature=0),
    graph=graph,
    verbose=True,
    return_intermediate_steps=True,
    allow_dangerous_requests=True,
)

result = chain.invoke({"query": "Who played in Top Gun?"})
print(f"Intermediate steps: {result['intermediate_steps']}")
print(f"Final answer: {result['result']}")

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE m.name = 'Top Gun'
RETURN a.name
Full Context:
[{'a.name': 'Tom Cruise'}, {'a.name': 'Val Kilmer'}, {'a.name': 'Anthony Edwards'}, {'a.name': 'Meg Ryan'}]

> Finished chain.
Intermediate steps: [{'query': "MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)\nWHERE m.name = 'Top Gun'\nRETURN a.name"}, {'context': [{'a.name': 'Tom Cruise'}, {'a.name': 'Val Kilmer'}, {'a.name': 'Anthony Edwards'}, {'a.name': 'Meg Ryan'}]}]
Final answer: Tom Cruise, Val Kilmer, Anthony Edwards, and Meg Ryan played in Top Gun.

​
返回直接结果

你可以使用 return_direct 参数返回 Cypher QA 链的直接结果。 
chain = GraphCypherQAChain.from_llm(
    ChatOpenAI(temperature=0),
    graph=graph,
    verbose=True,
    return_direct=True,
    allow_dangerous_requests=True,
)

chain.invoke({"query": "Who played in Top Gun?"})

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE m.name = 'Top Gun'
RETURN a.name

> Finished chain.

{'query': 'Who played in Top Gun?',
 'result': [{'a.name': 'Tom Cruise'},
  {'a.name': 'Val Kilmer'},
  {'a.name': 'Anthony Edwards'},
  {'a.name': 'Meg Ryan'}]}

​
在 Cypher 生成提示词中添加示例

你可以定义针对特定问题希望 LLM 生成的 Cypher 语句。 
from langchain_core.prompts.prompt import PromptTemplate

CYPHER_GENERATION_TEMPLATE = """Task:Generate Cypher statement to query a graph database.
Instructions:
Use only the provided relationship types and properties in the schema.
Do not use any other relationship types or properties that are not provided.
Schema:
{schema}
Note: Do not include any explanations or apologies in your responses.
Do not respond to any questions that might ask anything else than for you to construct a Cypher statement.
Do not include any text except the generated Cypher statement.
Examples: Here are a few examples of generated Cypher statements for particular questions:
# How many people played in Top Gun?
MATCH (m:Movie {{name:"Top Gun"}})<-[:ACTED_IN]-()
RETURN count(*) AS numberOfActors

The question is:
{question}"""

CYPHER_GENERATION_PROMPT = PromptTemplate(
    input_variables=["schema", "question"], template=CYPHER_GENERATION_TEMPLATE
)

chain = GraphCypherQAChain.from_llm(
    ChatOpenAI(temperature=0),
    graph=graph,
    verbose=True,
    cypher_prompt=CYPHER_GENERATION_PROMPT,
    allow_dangerous_requests=True,
)

chain.invoke({"query": "How many people played in Top Gun?"})

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (m:Movie {name:"Top Gun"})<-[:ACTED_IN]-()
RETURN count(*) AS numberOfActors
Full Context:
[{'numberOfActors': 4}]

> Finished chain.

{'query': 'How many people played in Top Gun?',
 'result': 'There were 4 actors in Top Gun.'}

​
为 Cypher 生成和答案生成使用独立的 LLM

你可以使用 cypher_llm 和 qa_llm 参数来定义不同的 LLM。 
chain = GraphCypherQAChain.from_llm(
    graph=graph,
    cypher_llm=ChatOpenAI(temperature=0, model="gpt-3.5-turbo"),
    qa_llm=ChatOpenAI(temperature=0, model="gpt-3.5-turbo-16k"),
    verbose=True,
    allow_dangerous_requests=True,
)

chain.invoke({"query": "Who played in Top Gun?"})

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE m.name = 'Top Gun'
RETURN a.name
Full Context:
[{'a.name': 'Tom Cruise'}, {'a.name': 'Val Kilmer'}, {'a.name': 'Anthony Edwards'}, {'a.name': 'Meg Ryan'}]

> Finished chain.

{'query': 'Who played in Top Gun?',
 'result': 'Tom Cruise, Val Kilmer, Anthony Edwards, and Meg Ryan played in Top Gun.'}

​
忽略指定的节点和关系类型

你可以在生成 Cypher 语句时使用 include_types 或 exclude_types 忽略图模式的某些部分。 
chain = GraphCypherQAChain.from_llm(
    graph=graph,
    cypher_llm=ChatOpenAI(temperature=0, model="gpt-3.5-turbo"),
    qa_llm=ChatOpenAI(temperature=0, model="gpt-3.5-turbo-16k"),
    verbose=True,
    exclude_types=["Movie"],
    allow_dangerous_requests=True,
)

# Inspect graph schema
print(chain.graph_schema)

Node properties are the following:
Actor {name: STRING}
Relationship properties are the following:

The relationships are the following:

​
验证生成的 Cypher 语句

你可以使用 validate_cypher 参数来验证并修正生成 Cypher 语句中的关系方向。 
chain = GraphCypherQAChain.from_llm(
    llm=ChatOpenAI(temperature=0, model="gpt-3.5-turbo"),
    graph=graph,
    verbose=True,
    validate_cypher=True,
    allow_dangerous_requests=True,
)

chain.invoke({"query": "Who played in Top Gun?"})

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE m.name = 'Top Gun'
RETURN a.name
Full Context:
[{'a.name': 'Tom Cruise'}, {'a.name': 'Val Kilmer'}, {'a.name': 'Anthony Edwards'}, {'a.name': 'Meg Ryan'}]

> Finished chain.

{'query': 'Who played in Top Gun?',
 'result': 'Tom Cruise, Val Kilmer, Anthony Edwards, and Meg Ryan played in Top Gun.'}

​
将数据库结果的上下文作为工具/函数输出提供

你可以使用 use_function_response 参数将数据库结果的上下文作为工具/函数输出传递给 LLM。该方法通过使 LLM 更紧密地遵循提供的上下文，提高了回答的准确性和相关性。 要使用此功能，你需要使用支持原生函数调用的 LLM。 
chain = GraphCypherQAChain.from_llm(
    llm=ChatOpenAI(temperature=0, model="gpt-3.5-turbo"),
    graph=graph,
    verbose=True,
    use_function_response=True,
    allow_dangerous_requests=True,
)
chain.invoke({"query": "Who played in Top Gun?"})

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE m.name = 'Top Gun'
RETURN a.name
Full Context:
[{'a.name': 'Tom Cruise'}, {'a.name': 'Val Kilmer'}, {'a.name': 'Anthony Edwards'}, {'a.name': 'Meg Ryan'}]

> Finished chain.

{'query': 'Who played in Top Gun?',
 'result': 'The main actors in Top Gun are Tom Cruise, Val Kilmer, Anthony Edwards, and Meg Ryan.'}
在使用函数响应功能时，你可以通过提供 function_response_system 来自定义系统消息，以指导模型如何生成答案。 请注意，使用 use_function_response 时 qa_prompt 将不起作用。 
chain = GraphCypherQAChain.from_llm(
    llm=ChatOpenAI(temperature=0, model="gpt-3.5-turbo"),
    graph=graph,
    verbose=True,
    use_function_response=True,
    function_response_system="Respond as a pirate!",
    allow_dangerous_requests=True,
)
chain.invoke({"query": "Who played in Top Gun?"})

> Entering new GraphCypherQAChain chain...
Generated Cypher:
MATCH (a:Actor)-[:ACTED_IN]->(m:Movie)
WHERE m.name = 'Top Gun'
RETURN a.name
Full Context:
[{'a.name': 'Tom Cruise'}, {'a.name': 'Val Kilmer'}, {'a.name': 'Anthony Edwards'}, {'a.name': 'Meg Ryan'}]

> Finished chain.

{'query': 'Who played in Top Gun?',
 'result': "Arrr matey! In the film Top Gun, ye be seein' Tom Cruise, Val Kilmer, Anthony Edwards, and Meg Ryan sailin' the high seas of the sky! Aye, they be a fine crew of actors, they be!"}
在 GitHub 上编辑此页面 或提交问题。
通过这些文档 经由 MCP 连接到 Claude、VSCode 等，以获取实时答案。