标题：基于 Google Gemini 和知识图谱实现的 QA 系统

! mkdir ad && cd ad

! curl https://arxiv.org/pdf/2106.07178.pdf --output AD1.pdf

! mv \*.pdf ad/

! pip install -q transformers

%pip install llama\_index pyvis Ipython langchain pypdf  llama-index-llms-huggingface  llama-index-embeddings-langchain llama-index-embeddings-huggingface

%pip install --upgrade --quiet llama-index-llms-gemini  google-generativeai

%pip install --upgrade --quiet llama-index-graph-stores-nebula nebulagraph-lite

import os



from google.colab import userdata

GOOGLE\_API\_KEY = userdata.get('GOOGLE\_API\_KEY')

os.environ["GOOGLE\_API\_KEY"] = GOOGLE\_API\_KEY

import logging

import sys



logging.basicConfig(stream=sys.stdout, level=logging.INFO)

logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))

from llama\_index.core import (

                         ServiceContext,

                         KnowledgeGraphIndex)

from llama\_index.core import SimpleDirectoryReader

from llama\_index.core.storage.storage\_context import StorageContext

from pyvis.network import Network



from llama\_index.llms.huggingface import HuggingFaceLLM

import google.generativeai as genai





for m in genai.list\_models():

    if "generateContent" in m.supported\_generation\_methods:

        print(m.name)

        print(m)

from llama\_index.llms.gemini import Gemini



llm = Gemini(model="models/gemini-1.0-pro-latest")

from llama\_index.embeddings.huggingface import HuggingFaceEmbedding

from llama\_index.core import ServiceContext





embed\_model = HuggingFaceEmbedding(model\_name="BAAI/bge-small-en-v1.5")

documents = SimpleDirectoryReader("/content/ad").load\_data()

print(len(documents))

from nebulagraph\_lite import nebulagraph\_let

n = nebulagraph\_let(debug=False)

n.start()

%ngql --address 127.0.0.1 --port 9669 --user root --password nebula

# If not, create it with the following commands from NebulaGraph's console:

%ngql CREATE SPACE nebula\_ad(vid\_type=FIXED\_STRING(256), partition\_num=1, replica\_factor=1)

import time



print("Waiting...")



# Delay for 10 seconds

time.sleep(10)



%ngql --address 127.0.0.1 --port 9669 --user root --password nebula

%ngql USE nebula\_ad;

%ngql CREATE TAG entity(name string);

%ngql CREATE EDGE relationship(relationship string);

import os

os.environ["NEBULA\_USER"] = "root"

os.environ["NEBULA\_PASSWORD"] = "nebula"  # default is "nebula"

os.environ[

    "NEBULA\_ADDRESS"

] = "127.0.0.1:9669"  # assumed we have NebulaGraph installed locally



space\_name = "nebula\_ad"

edge\_types, rel\_prop\_names = ["relationship"], [

    "relationship"

]  # default, could be omit if create from an empty kg

tags = ["entity"]  # default, could be omit if create from an empty kg

from llama\_index.core import StorageContext

from llama\_index.graph\_stores.nebula import NebulaGraphStore



graph\_store = NebulaGraphStore(

    space\_name=space\_name,

    edge\_types=edge\_types,

    rel\_prop\_names=rel\_prop\_names,

    tags=tags,

)

storage\_context = StorageContext.from\_defaults(graph\_store=graph\_store)

from llama\_index.core import Settings



Settings.llm = llm

Settings.embed\_model = embed\_model

Settings.chunk\_size = 512

# NOTE: can take a while!

index = KnowledgeGraphIndex.from\_documents(

    documents,

    storage\_context=storage\_context,

    max\_triplets\_per\_chunk=10,

    space\_name=space\_name,

    edge\_types=edge\_types,

    rel\_prop\_names=rel\_prop\_names,

    tags=tags,

    include\_embeddings=True

)

# Query some random Relationships with Cypher

%ngql USE nebula\_ad;

%ngql MATCH ()-[e]->() RETURN e LIMIT 10

query\_engine = index.as\_query\_engine()

from IPython.display import display, Markdown



response = query\_engine.query(

    "Tell me about Anomaly?",

)

display(Markdown(f"<b>{response}</b>"))

from IPython.display import display, Markdown



response = query\_engine.query(

    "What are graph anomolies?",

)

display(Markdown(f"<b>{response}</b>"))