URI: https://demo.neo4jlabs.com:7473/browser/
用戶名: companies
密碼: companies
數(shù)據(jù)庫: companies

import os

os.environ["OPENAI_API_KEY"] = "sk-"
os.environ["NEO4J_URI"] = "neo4j+s://demo.neo4jlabs.com"
os.environ["NEO4J_USERNAME"] = "companies"
os.environ["NEO4J_PASSWORD"] = "companies"
os.environ["NEO4J_DATABASE"] = "companies"

embeddings = OpenAIEmbeddings()
graph = Neo4jGraph()
vector_index = Neo4jVector.from_existing_index(
 embeddings,
 index_name="news"
)

def get_organization_news(
 topic: Optional[str] = None,
 organization: Optional[str] = None,
 country: Optional[str] = None,
 sentiment: Optional[str] = None,
) -> str:
 # 如果沒有預(yù)過濾條件，我們可以直接使用向量索引進(jìn)行搜索
 if topic and not organization and not country and not sentiment:
     return vector_index.similarity_search(topic)
 # 使用并行運(yùn)行時(shí)（如果可用）
 base_query = (
     "CYPHER runtime = parallel parallelRuntimeSupport=all "
     "MATCH (c:Chunk)<-[:HAS_CHUNK]-(a:Article) WHERE "
 )
 where_queries = []
 params = {"k": 5}  # 設(shè)置要檢索的文本塊數(shù)量

if organization:
    # 將組織名稱映射到數(shù)據(jù)庫中的候選項(xiàng)
    candidates = get_candidates(organization)
    if len(candidates) > 1:  # 如果候選選項(xiàng)太多，則需要用戶進(jìn)一步明確
        return f"請(qǐng)明確指出用戶指的是以下哪個(gè)組織：{candidates}"
    # 添加一個(gè)過濾條件，篩選出提及特定組織的 articles
    where_queries.append(f"EXISTS {{(a)-[:MENTIONS]->(:Organization {{name: $organization}})}}")
    # 將組織名稱作為參數(shù)傳入
    params["organization"] = candidates[0]

if country:
    # 由于國家名稱標(biāo)準(zhǔn)化，不需要額外的映射
    where_queries.append(f"EXISTS {{(a)-[:MENTIONS]->(:Organization)-[:IN_CITY]->()-[:IN_COUNTRY]->(:Country {{name: $country}})}}")
    params["country"] = country

if sentiment:
    if sentiment == "positive":
        where_queries.append("a.sentiment > $sentiment")
        params["sentiment"] = 0.5
    else:
        where_queries.append("a.sentiment < $sentiment")
        params["sentiment"] = -0.5

if topic:  # 執(zhí)行向量比較
    vector_snippet = (
        "WITH c, a, vector.similarity.cosine(c.embedding,$embedding) AS score "
        "ORDER BY score DESC LIMIT toInteger($k)"
    )
    params["embedding"] = embeddings.embed_query(topic)
else:  # 只返回最新的數(shù)據(jù)
    vector_snippet = "WITH c, a ORDER BY a.date DESC LIMIT toInteger($k)"

return_snippet = "RETURN '#title ' + a.title + '\n#date ' + toString(a.date) + '\n#text ' + c.text AS output"

complete_query = (
    base_query + " AND ".join(where_queries) + vector_snippet + return_snippet
)

# 從數(shù)據(jù)庫檢索信息
data = graph.query(complete_query, params)
print(f"Cypher: {complete_query}\n")
# 在打印前安全地移除嵌入
params.pop('embedding', None)
print(f"參數(shù): {params}")
return "###文章: ".join([el["output"] for el in data])

get_organization_news(
  organizatinotallow='neo4j',
  sentiment='positive',
  topic='遠(yuǎn)程工作'
)

# Cypher: CYPHER runtime = parallel parallelRuntimeSupport=all
# MATCH (c:Chunk)<-[:HAS_CHUNK]-(a:Article) WHERE 
# EXISTS {(a)-[:MENTIONS]->(:Organization {name: $organization})} AND 
# a.sentiment > $sentiment 
# WITH c, a, vector.similarity.cosine(c.embedding,$embedding) AS score 
# ORDER BY score DESC LIMIT toInteger($k) 
# RETURN '#title ' + a.title + '\n#date ' + toString(a.date) + '\n#text ' + c.text AS output

# 參數(shù): {'k': 5, 'organization': 'Neo4j', 'sentiment': 0.5}

fewshot_examples = """{輸入：Google員工的健康福利在新聞中有哪些？查詢：健康福利}
{輸入：關(guān)于Google的最新正面新聞是什么？查詢：無}
{輸入：有關(guān)VertexAI和Google的新聞?dòng)心男坎樵儯篤ertexAI}
{輸入：關(guān)于Google的新產(chǎn)品有哪些新聞？查詢：新產(chǎn)品}
"""

class NewsInput(BaseModel):
    topic: Optional[str] = Field(
        descriptinotallow="除了組織、國家和情感傾向之外，如果您對(duì)其他特定信息或話題感興趣，請(qǐng)告訴我們。以下是一些示例："
        + fewshot_examples
    )
    organization: Optional[str] = Field(
        descriptinotallow="您希望了解信息的組織名稱"
    )
    country: Optional[str] = Field(
        descriptinotallow="您感興趣的組織的所在國家。請(qǐng)使用正式的國家名稱，例如‘美利堅(jiān)合眾國’或‘法國’。"
    )
    sentiment: Optional[str] = Field(
        descriptinotallow="您想要查詢的文章情感傾向", enum=["正面", "負(fù)面"]
    )

class NewsTool(BaseTool):
    name = "新聞信息工具"
    description = (
        "當(dāng)你需要在新聞中查找相關(guān)信息時(shí)，這個(gè)工具會(huì)非常有用。"
    )
    args_schema:Type[BaseModel] = NewsInput

    def _run(
        self,
        topic: Optional[str] = None,
        organization: Optional[str] = None,
        country: Optional[str] = None,
        sentiment: Optional[str] = None,
        run_manager: Optional[CallbackManagerForToolRun] = None,
    ) -> str:
        ""“使用這個(gè)工具來獲取新聞信息。”""
        return get_organization_news(topic, organization, country, sentiment)

llm = ChatOpenAI(temperature=0, model="gpt-4-turbo", streaming=True)
tools = [NewsTool()]

llm_with_tools = llm.bind(functinotallow=[format_tool_to_openai_function(t) for t in tools])

prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            “你是一個(gè)樂于助人的助手，可以找到關(guān)于電影的信息并進(jìn)行推薦。如果工具需要進(jìn)一步的問題，請(qǐng)確保向用戶詢問以獲得澄清。確保在后續(xù)問題中包含任何需要澄清的可用選項(xiàng)。只做用戶明確請(qǐng)求的事情。”
        ),
        MessagesPlaceholder(variable_name="chat_history"),
        ("user", "{input}"),
        MessagesPlaceholder(variable_name="agent_scratchpad"),
    ]
)

agent = (
    {
        "input": lambda x: x["input"],
        "chat_history": lambda x: _format_chat_history(x["chat_history"])
        if x.get("chat_history")
        else [],
        "agent_scratchpad": lambda x: format_to_openai_function_messages(
            x["intermediate_steps"]
        ),
    }
    | prompt
    | llm_with_tools
    | OpenAIFunctionsAgentOutputParser()
)

agent_executor = AgentExecutor(agent=agent, tools=tools)

agent_executor.invoke(
  {"輸入": "關(guān)于 neo4j 的一些正面新聞是什么？"}
)

# Cypher: CYPHER runtime = parallel parallelRuntimeSupport=all 
# MATCH (c:Chunk)<-[:HAS_CHUNK]-(a:Article) WHERE 
# EXISTS {(a)-[:MENTIONS]->(:Organization {name: $organization})} AND 
# a.sentiment > $sentiment WITH c, a 
# ORDER BY a.date DESC LIMIT toInteger($k) 
# RETURN '#標(biāo)題 ' + a.title + '日期 ' + toString(a.date) + '文本 ' + c.text AS output
# 參數(shù): {'k': 5, 'organization': 'Neo4j', 'sentiment': 0.5}

agent_executor.invoke(
   {"輸入": "關(guān)于法國公司的員工幸福感，有哪些最新的負(fù)面新聞？"}
)

# Cypher: CYPHER runtime = parallel parallelRuntimeSupport=all 
# MATCH (c:Chunk)<-[:HAS_CHUNK]-(a:Article) WHERE 
# EXISTS {(a)-[:MENTIONS]->(:Organization)-[:IN_CITY]->()-[:IN_COUNTRY]->(:Country {name: $country})} AND 
# a.sentiment < $sentiment 
# WITH c, a, vector.similarity.cosine(c.embedding,$embedding) AS score 
# ORDER BY score DESC LIMIT toInteger($k) 
# RETURN '#標(biāo)題 ' + a.title + '日期 ' + toString(a.date) + '文本 ' + c.text AS output
# 參數(shù): {'k': 5, 'country': 'France', 'sentiment': -0.5, 'topic': '員工幸福感'}