FastGPT 引申：信息抽取到知识图谱的衔接流程

关于知识图谱构建中信息抽取结果与Cypher代码的衔接逻辑，这里通过一个完整的流程演示说明：

信息抽取到知识图谱的衔接流程

步骤1：原始信息抽取结果

{
  "athlete": "Yusuf Dikeç",
  "nationality": "Turkey",
  "event": "10m Air Pistol",
  "medal": "Silver",
  "game": {"year":2024, "location":"Paris"},
  "score": 243.7
}

步骤2：数据标准化处理（Python示例）

# 将抽取结果转换为图谱节点/关系模板
def convert_to_graph_data(extracted_data):
    return {
        "athlete": {
            "id": f"ATH_{extracted_data['nationality']}_001",
            "name": extracted_data["athlete"],
            "nationality": extracted_data["nationality"]
        },
        "event": {
            "id": "EVT_10MAP",
            "name": extracted_data["event"],
            "discipline": "Shooting"
        },
        "relationship": {
            "type": "WON_MEDAL",
            "properties": {
                "type": extracted_data["medal"],
                "score": extracted_data["score"]
            }
        }
    }

# 输出结构化图谱数据
graph_data = convert_to_graph_data(extracted_data)
"""
{
  "athlete": {"id": "ATH_Turkey_001", "name": "Yusuf Dikeç", ...},
  "event": {"id": "EVT_10MAP", "name": "10m Air Pistol", ...},
  "relationship": {"type": "WON_MEDAL", "properties": {...}}
}
"""

步骤3：Cypher代码动态生成（Python驱动）

from neo4j import GraphDatabase

class Neo4jLoader:
    def __init__(self, uri, user, password):
        self.driver = GraphDatabase.driver(uri, auth=(user, password))
        
    def create_relationship(self, graph_data):
        with self.driver.session() as session:
            # 节点创建（使用MERGE防止重复）
            session.run("""
                MERGE (a:Athlete {id: $a_id}) 
                SET a.name = $a_name, a.nationality = $a_nationality
                MERGE (e:Event {id: $e_id})
                SET e.name = $e_name, e.discipline = $e_discipline
                """, 
                a_id=graph_data["athlete"]["id"],
                a_name=graph_data["athlete"]["name"],
                a_nationality=graph_data["athlete"]["nationality"],
                e_id=graph_data["event"]["id"],
                e_name=graph_data["event"]["name"],
                e_discipline=graph_data["event"]["discipline"]
            )
            
            # 关系创建
            session.run("""
                MATCH (a:Athlete {id: $a_id}), (e:Event {id: $e_id})
                CREATE (a)-[r:WON_MEDAL]->(e)
                SET r += $props
                """,
                a_id=graph_data["athlete"]["id"],
                e_id=graph_data["event"]["id"],
                props=graph_data["relationship"]["properties"]
            )

# 使用示例
loader = Neo4jLoader("bolt://localhost:7687", "neo4j", "password")
loader.create_relationship(graph_data)

关键衔接逻辑说明

1. 唯一标识符生成规则

# 运动员ID生成逻辑
f"ATH_{nationality_code}_{sequence_num}"  # 示例: ATH_Turkey_001

# 赛事ID生成逻辑
f"EVT_{discipline_code}{event_code}"     # 示例: EVT_10MAP (10m Air Pistol)

2. 数据映射策略

3. 批处理场景示例

# 当有多个运动员数据时
batch_data = [graph_data1, graph_data2, graph_data3]

for data in batch_data:
    # 自动生成带序列号的ID
    data["athlete"]["id"] = generate_athlete_id(data["nationality"], seq_num) 
    
    # 执行节点和关系创建
    loader.create_relationship(data)

4. 冲突处理机制

// 使用MERGE+ON CREATE保证幂等性
MERGE (a:Athlete {id: $a_id}) 
ON CREATE SET a.createTime = timestamp()
ON MATCH SET a.updateTime = timestamp()

// 关系存在性检查
OPTIONAL MATCH (a)-[r:WON_MEDAL]->(e) 
WHERE r.score < $new_score
DELETE r
CREATE (a)-[r_new:WON_MEDAL]->(e)

可视化衔接流程

原始文本 → 信息抽取 → 标准化JSON → Cypher模板填充 → 图数据库写入
           (Mistral-7B)       ↑            ↓
                              数据校验 ← 类型转换

通过这种方式，信息抽取结果中的非结构化数据被系统地转化为知识图谱中的节点、属性和关系，同时保证了数据的一致性和可追溯性。