知识图谱组件

本章节详细介绍了 HRAG 系统中的知识图谱组件。

知识图谱构建提供 HiRAG 与 TRAG 两种方法。两种方法均由以下三部分组成：

  1. 构建实体关系三元组

  2. 生成实体关系对应描述

  3. 构建知识图谱

其中两种方法共用同一个构建实体关系三元组的方法。对应的参数设置见 knowledge_graph 相关参数配置指南

从语料中抽取实体关系：

提供两种抽取实体关系的方法

  1. ConmmonKG方法，使用代码：

from src.data_processor.knowledge_graph.entity_relation_extractor import entity_relation_extractor


# 根据MinerU生成的文件得到三元组
mineru_path = "src/resources/pdf"
output_path = "src/resources/temp/knowledge_graph/commonkg"
entity_relation_extractor(mineru_path, output_path, corpus_dir = corpus_path, method="CommonKG")

  1. GraphRAG方法，使用代码：

from src.data_processor.knowledge_graph.entity_relation_extractor import entity_relation_extractor


# 根据MinerU生成的文件得到三元组
mineru_path = "src/resources/pdf"
output_path = "src/resources/temp/knowledge_graph/graphrag"
entity_relation_extractor(mineru_path, output_path, method="graphrag")

构建知识图谱：

# HiRAG
from src.data_processor.knowledge_graph.graph_builder import graph_builder

# 实体关系三元组等数据构建hirag，并存入working_dir

# 选择合适的实体关系提取方法
data_path = "src/resources/temp/knowledge_graph/graphrag" # or "src/resources/temp/knowledge_graph/commonkg"
working_dir = "src/resources/temp/knowledge_graph/hirag"
graph_builder(data_path, working_dir,method="hirag")

# LeanRAG
from src.data_processor.knowledge_graph.graph_builder import graph_builder

# 实体关系三元组等数据构建leanrag，并存入working_dir
data_path = "src/resources/temp/knowledge_graph/trag_data"
working_dir = "src/resources/temp/knowledge_graph/leanrag"
graph_builder(data_path, working_dir,method="leanrag")

查询知识图谱：

# HiRAG
from src.data_processor.knowledge_graph.query_graph import query_graph

query = "Which leadership positions changed at Datalogic in the reporting period?"
working_dir = "src/resources/temp/knowledge_graph/hirag"
result = query_graph(query, working_dir, method="hirag")
print(result)

# LeanRAG
from src.data_processor.knowledge_graph.query_graph import query_graph

query = "Which leadership positions changed at Datalogic in the reporting period?"
working_dir = "src/resources/temp/knowledge_graph/leanrag"
result = query_graph(query, working_dir, method="leanrag")
print(result)