Prompting strategies
In this guide we'll go over prompting strategies to improve graph database query generation. We'll largely focus on methods for getting relevant database-specific information in your prompt.
Setup
First, get required packages and set environment variables:
%pip install --upgrade --quiet langchain langchain-community langchain-openai neo4j
Note: you may need to restart the kernel to use updated packages.
We default to OpenAI models in this guide, but you can swap them out for the model provider of your choice.
import getpass
import os
os.environ["OPENAI_API_KEY"] = getpass.getpass()
# Uncomment the below to use LangSmith. Not required.
# os.environ["LANGCHAIN_API_KEY"] = getpass.getpass()
# os.environ["LANGCHAIN_TRACING_V2"] = "true"
········
Next, we need to define Neo4j credentials. Follow these installation steps to set up a Neo4j database.
os.environ["NEO4J_URI"] = "bolt://localhost:7687"
os.environ["NEO4J_USERNAME"] = "neo4j"
os.environ["NEO4J_PASSWORD"] = "password"
The below example will create a connection with a Neo4j database and will populate it with example data about movies and their actors.
from langchain_community.graphs import Neo4jGraph
graph = Neo4jGraph()
# Import movie information
movies_query = """
LOAD CSV WITH HEADERS FROM
'https://raw.githubusercontent.com/tomasonjo/blog-datasets/main/movies/movies_small.csv'
AS row
MERGE (m:Movie {id:row.movieId})
SET m.released = date(row.released),
m.title = row.title,
m.imdbRating = toFloat(row.imdbRating)
FOREACH (director in split(row.director, '|') |
MERGE (p:Person {name:trim(director)})
MERGE (p)-[:DIRECTED]->(m))
FOREACH (actor in split(row.actors, '|') |
MERGE (p:Person {name:trim(actor)})
MERGE (p)-[:ACTED_IN]->(m))
FOREACH (genre in split(row.genres, '|') |
MERGE (g:Genre {name:trim(genre)})
MERGE (m)-[:IN_GENRE]->(g))
"""
graph.query(movies_query)
API Reference:
[]
Filtering graph schema
At times, you may need to focus on a specific subset of the graph schema while generating Cypher statements. Let's say we are dealing with the following graph schema:
graph.refresh_schema()
print(graph.schema)
Node properties are the following:
Movie {imdbRating: FLOAT, id: STRING, released: DATE, title: STRING},Person {name: STRING},Genre {name: STRING}
Relationship properties are the following:
The relationships are the following:
(:Movie)-[:IN_GENRE]->(:Genre),(:Person)-[:DIRECTED]->(:Movie),(:Person)-[:ACTED_IN]->(:Movie)
Let's say we want to exclude the Genre node from the schema representation we pass to an LLM.
We can achieve that using the exclude parameter of the GraphCypherQAChain chain.
from langchain.chains import GraphCypherQAChain
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)
chain = GraphCypherQAChain.from_llm(
graph=graph, llm=llm, exclude_types=["Genre"], verbose=True
)
API Reference:
print(chain.graph_schema)
Node properties are the following:
Movie {imdbRating: FLOAT, id: STRING, released: DATE, title: STRING},Person {name: STRING}
Relationship properties are the following:
The relationships are the following:
(:Person)-[:DIRECTED]->(:Movie),(:Person)-[:ACTED_IN]->(:Movie)
Few-shot examples
Including examples of natural language questions being converted to valid Cypher queries against our database in the prompt will often improve model performance, especially for complex queries.
Let's say we have the following examples:
examples = [
{
"question": "How many artists are there?",
"query": "MATCH (a:Person)-[:ACTED_IN]->(:Movie) RETURN count(DISTINCT a)",
},
{
"question": "Which actors played in the movie Casino?",
"query": "MATCH (m:Movie {{title: 'Casino'}})<-[:ACTED_IN]-(a) RETURN a.name",
},
{
"question": "How many movies has Tom Hanks acted in?",
"query": "MATCH (a:Person {{name: 'Tom Hanks'}})-[:ACTED_IN]->(m:Movie) RETURN count(m)",
},
{
"question": "List all the genres of the movie Schindler's List",
"query": "MATCH (m:Movie {{title: 'Schindler\\'s List'}})-[:IN_GENRE]->(g:Genre) RETURN g.name",
},
{
"question": "Which actors have worked in movies from both the comedy and action genres?",
"query": "MATCH (a:Person)-[:ACTED_IN]->(:Movie)-[:IN_GENRE]->(g1:Genre), (a)-[:ACTED_IN]->(:Movie)-[:IN_GENRE]->(g2:Genre) WHERE g1.name = 'Comedy' AND g2.name = 'Action' RETURN DISTINCT a.name",
},
{
"question": "Which directors have made movies with at least three different actors named 'John'?",
"query": "MATCH (d:Person)-[:DIRECTED]->(m:Movie)<-[:ACTED_IN]-(a:Person) WHERE a.name STARTS WITH 'John' WITH d, COUNT(DISTINCT a) AS JohnsCount WHERE JohnsCount >= 3 RETURN d.name",
},
{
"question": "Identify movies where directors also played a role in the film.",
"query": "MATCH (p:Person)-[:DIRECTED]->(m:Movie), (p)-[:ACTED_IN]->(m) RETURN m.title, p.name",
},
{
"question": "Find the actor with the highest number of movies in the database.",
"query": "MATCH (a:Actor)-[:ACTED_IN]->(m:Movie) RETURN a.name, COUNT(m) AS movieCount ORDER BY movieCount DESC LIMIT 1",
},
]
We can create a few-shot prompt with them like so:
from langchain_core.prompts import FewShotPromptTemplate, PromptTemplate
example_prompt = PromptTemplate.from_template(
"User input: {question}\nCypher query: {query}"
)
prompt = FewShotPromptTemplate(
examples=examples[:5],
example_prompt=example_prompt,
prefix="You are a Neo4j expert. Given an input question, create a syntactically correct Cypher query to run.\n\nHere is the schema information\n{schema}.\n\nBelow are a number of examples of questions and their corresponding Cypher queries.",
suffix="User input: {question}\nCypher query: ",
input_variables=["question", "schema"],
)
API Reference:
print(prompt.format(question="How many artists are there?", schema="foo"))
You are a Neo4j expert. Given an input question, create a syntactically correct Cypher query to run.
Here is the schema information
foo.
Below are a number of examples of questions and their corresponding Cypher queries.
User input: How many artists are there?
Cypher query: MATCH (a:Person)-[:ACTED_IN]->(:Movie) RETURN count(DISTINCT a)
User input: Which actors played in the movie Casino?
Cypher query: MATCH (m:Movie {title: 'Casino'})<-[:ACTED_IN]-(a) RETURN a.name
User input: How many movies has Tom Hanks acted in?
Cypher query: MATCH (a:Person {name: 'Tom Hanks'})-[:ACTED_IN]->(m:Movie) RETURN count(m)
User input: List all the genres of the movie Schindler's List
Cypher query: MATCH (m:Movie {title: 'Schindler\'s List'})-[:IN_GENRE]->(g:Genre) RETURN g.name
User input: Which actors have worked in movies from both the comedy and action genres?
Cypher query: MATCH (a:Person)-[:ACTED_IN]->(:Movie)-[:IN_GENRE]->(g1:Genre), (a)-[:ACTED_IN]->(:Movie)-[:IN_GENRE]->(g2:Genre) WHERE g1.name = 'Comedy' AND g2.name = 'Action' RETURN DISTINCT a.name
User input: How many artists are there?
Cypher query:
Dynamic few-shot examples
If we have enough examples, we may want to only include the most relevant ones in the prompt, either because they don't fit in the model's context window or because the long tail of examples distracts the model. And specifically, given any input we want to include the examples most relevant to that input.
We can do just this using an ExampleSelector. In this case we'll use a SemanticSimilarityExampleSelector, which will store the examples in the vector database of our choosing. At runtime it will perform a similarity search between the input and our examples, and return the most semantically similar ones:
from langchain_community.vectorstores import Neo4jVector
from langchain_core.example_selectors import SemanticSimilarityExampleSelector
from langchain_openai import OpenAIEmbeddings
example_selector = SemanticSimilarityExampleSelector.from_examples(
examples,
OpenAIEmbeddings(),
Neo4jVector,
k=5,
input_keys=["question"],
)
example_selector.select_examples({"question": "how many artists are there?"})
[{'query': 'MATCH (a:Person)-[:ACTED_IN]->(:Movie) RETURN count(DISTINCT a)',
'question': 'How many artists are there?'},
{'query': "MATCH (a:Person {{name: 'Tom Hanks'}})-[:ACTED_IN]->(m:Movie) RETURN count(m)",
'question': 'How many movies has Tom Hanks acted in?'},
{'query': "MATCH (a:Person)-[:ACTED_IN]->(:Movie)-[:IN_GENRE]->(g1:Genre), (a)-[:ACTED_IN]->(:Movie)-[:IN_GENRE]->(g2:Genre) WHERE g1.name = 'Comedy' AND g2.name = 'Action' RETURN DISTINCT a.name",
'question': 'Which actors have worked in movies from both the comedy and action genres?'},
{'query': "MATCH (d:Person)-[:DIRECTED]->(m:Movie)<-[:ACTED_IN]-(a:Person) WHERE a.name STARTS WITH 'John' WITH d, COUNT(DISTINCT a) AS JohnsCount WHERE JohnsCount >= 3 RETURN d.name",
'question': "Which directors have made movies with at least three different actors named 'John'?"},
{'query': 'MATCH (a:Actor)-[:ACTED_IN]->(m:Movie) RETURN a.name, COUNT(m) AS movieCount ORDER BY movieCount DESC LIMIT 1',
'question': 'Find the actor with the highest number of movies in the database.'}]
To use it, we can pass the ExampleSelector directly in to our FewShotPromptTemplate:
prompt = FewShotPromptTemplate(
example_selector=example_selector,
example_prompt=example_prompt,
prefix="You are a Neo4j expert. Given an input question, create a syntactically correct Cypher query to run.\n\nHere is the schema information\n{schema}.\n\nBelow are a number of examples of questions and their corresponding Cypher queries.",
suffix="User input: {question}\nCypher query: ",
input_variables=["question", "schema"],
)
print(prompt.format(question="how many artists are there?", schema="foo"))
You are a Neo4j expert. Given an input question, create a syntactically correct Cypher query to run.
Here is the schema information
foo.
Below are a number of examples of questions and their corresponding Cypher queries.
User input: How many artists are there?
Cypher query: MATCH (a:Person)-[:ACTED_IN]->(:Movie) RETURN count(DISTINCT a)
User input: How many movies has Tom Hanks acted in?
Cypher query: MATCH (a:Person {name: 'Tom Hanks'})-[:ACTED_IN]->(m:Movie) RETURN count(m)
User input: Which actors have worked in movies from both the comedy and action genres?
Cypher query: MATCH (a:Person)-[:ACTED_IN]->(:Movie)-[:IN_GENRE]->(g1:Genre), (a)-[:ACTED_IN]->(:Movie)-[:IN_GENRE]->(g2:Genre) WHERE g1.name = 'Comedy' AND g2.name = 'Action' RETURN DISTINCT a.name
User input: Which directors have made movies with at least three different actors named 'John'?
Cypher query: MATCH (d:Person)-[:DIRECTED]->(m:Movie)<-[:ACTED_IN]-(a:Person) WHERE a.name STARTS WITH 'John' WITH d, COUNT(DISTINCT a) AS JohnsCount WHERE JohnsCount >= 3 RETURN d.name
User input: Find the actor with the highest number of movies in the database.
Cypher query: MATCH (a:Actor)-[:ACTED_IN]->(m:Movie) RETURN a.name, COUNT(m) AS movieCount ORDER BY movieCount DESC LIMIT 1
User input: how many artists are there?
Cypher query:
llm = ChatOpenAI(model="gpt-3.5-turbo", temperature=0)
chain = GraphCypherQAChain.from_llm(
graph=graph, llm=llm, cypher_prompt=prompt, verbose=True
)
chain.invoke("How many actors are in the graph?")
[1m> Entering new GraphCypherQAChain chain...[0m
Generated Cypher:
[32;1m[1;3mMATCH (a:Person)-[:ACTED_IN]->(:Movie) RETURN count(DISTINCT a)[0m
Full Context:
[32;1m[1;3m[{'count(DISTINCT a)': 967}][0m
[1m> Finished chain.[0m
{'query': 'How many actors are in the graph?',
'result': 'There are 967 actors in the graph.'}