AgensGraph 完全指南 / 第 07 章：图数据建模

第 07 章：图数据建模

7.1 图建模方法论

7.1.1 建模思维转变

从关系型建模到图建模，需要进行思维转换：

关系型建模思维:
  1. 识别实体 → 设计表
  2. 识别关系 → 外键
  3. 规范化 → 拆表消除冗余
  4. 反规范化 → 按需合并表

图建模思维:
  1. 识别实体 → 设计顶点（Vertex Label）
  2. 识别关系 → 设计边（Edge Label）
  3. 识别属性 → 分配到顶点和边上
  4. 设计遍历路径 → 优化高频查询路径

7.1.2 建模四步法

7.2 顶点设计

7.2.1 标签设计原则

-- ❌ 错误：过于细粒度的标签
CREATE (n:Person:Employee:Manager:FullTime:BeijingOffice {});

-- ✅ 正确：使用核心分类标签
CREATE (n:Person:Employee {name: '张三', role: 'manager', office: '北京'});

-- 标签设计原则：
-- 1. 标签表示"是什么"，属性表示"属于什么分类"
-- 2. 一般不超过 3 个标签
-- 3. 标签应有业务含义

7.2.2 标签命名规范

7.2.3 属性设计

-- 良好的属性设计
CREATE (:Person {
  name: '张三',                -- 必填属性
  age: 30,                    -- 基本属性
  email: 'zhangsan@example.com', -- 唯一标识属性
  skills: ['Java', 'Python'],   -- 数组属性
  address: {                  -- 嵌套属性
    city: '北京',
    district: '海淀'
  },
  created: datetime(),        -- 时间戳
  version: 1                  -- 版本号（乐观锁）
});

属性设计原则：

7.3 边设计

7.3.1 边的粒度

-- ❌ 过细：每种关系一种类型
(a)-[:KNOWS]->(b)
(a)-[:FRIENDS_WITH]->(b)
(a)-[:FOLLOWS]->(b)
(a)-[:COLLEAGUE_OF]->(b)

-- ✅ 合理：使用属性区分关系细节
(a)-[:RELATED_TO {type: 'friend', since: 2020}]->(b)
(a)-[:RELATED_TO {type: 'colleague', since: 2021}]->(b)

-- ✅ 最佳：有意义的关系类型 + 关键属性
(a)-[:KNOWS {context: '大学', since: 2018}]->(b)
(a)-[:WORKS_WITH {project: 'GraphDB', since: 2022}]->(b)

7.3.2 边的命名规范

7.3.3 何时将关系提升为顶点

当关系本身需要被关联时，将其"提升"（Reification）为顶点：

场景：一次购买包含多个产品

❌ 简单边方案:
  (Customer)-[:BOUGHT {quantity: 2, date: '2024-01-01'}]->(Product)

✅ 中间顶点方案:
  (Customer)-[:PLACED]->(Order)-[:CONTAINS]->(Product)
                         (Order)-[:SHIPPED_TO]->(Address)
                         (Order {date: datetime(), total: 99.99})

-- 创建订单中间顶点
MATCH (c:Customer {id: 'C001'}), (p:Product {id: 'P001'})
CREATE (c)-[:PLACED]->(o:Order {
  order_id: 'O001',
  date: datetime(),
  status: 'pending'
})-[:CONTAINS {quantity: 2, unit_price: 49.99}]->(p);

7.4 经典图模型模式

7.4.1 社交网络模型

社交网络图模型:

  (Person)-[:KNOWS {since, context}]->(Person)
  (Person)-[:FOLLOWS]->(Person)
  (Person)-[:MEMBER_OF]->(Group)
  (Person)-[:POSTED]->(Post)
  (Post)-[:TAGGED_WITH]->(Tag)
  (Person)-[:LIKED]->(Post)
  (Person)-[:COMMENTED_ON]->(Post)

-- 社交网络 Schema
CREATE (:Person {id, name, email, bio, joined});
CREATE (:Group {id, name, description, created});
CREATE (:Post {id, content, created, likes_count});
CREATE (:Tag {name});

-- 关系类型
(:Person)-[:KNOWS {since, context}]->(:Person)
(:Person)-[:FOLLOWS]->(:Person)
(:Person)-[:MEMBER_OF {role, since}]->(:Group)
(:Person)-[:POSTED]->(:Post)
(:Post)-[:TAGGED_WITH]->(:Tag)
(:Person)-[:LIKED {at}]->(:Post)

7.4.2 企业组织模型

企业组织图:

  (Employee)-[:REPORTS_TO]->(Employee)
  (Employee)-[:BELONGS_TO]->(Department)
  (Department)-[:PART_OF]->(Department)
  (Employee)-[:HAS_SKILL]->(Skill)
  (Employee)-[:WORKS_ON]->(Project)
  (Project)-[:DELIVERS]->(Product)

-- 企业组织 Schema
CREATE (:Employee {id, name, title, level, salary, hire_date});
CREATE (:Department {id, name, budget, location});
CREATE (:Project {id, name, start_date, end_date, status});
CREATE (:Skill {name, category, level});

-- 关系
(:Employee)-[:REPORTS_TO {since}]->(:Employee)
(:Employee)-[:BELONGS_TO {role, since}]->(:Department)
(:Department)-[:PART_OF]->(:Department)
(:Employee)-[:HAS_SKILL {proficiency, years}]->(:Skill)
(:Employee)-[:WORKS_ON {role, allocation}]->(:Project)

7.4.3 知识图谱模型

知识图谱:

  (Entity)-[:IS_A]->(Type)
  (Entity)-[:RELATED_TO {relation}]->(Entity)
  (Entity)-[:HAS_PROPERTY]->(PropertyValue)
  (Concept)-[:SUBCLASS_OF]->(Concept)
  (Document)-[:MENTIONS]->(Entity)

-- 知识图谱通用 Schema
CREATE (:Entity {id, name, description, source});
CREATE (:Type {name, namespace});
CREATE (:Document {id, title, content, url});
CREATE (:Concept {name, definition});

(:Entity)-[:IS_A]->(:Type)
(:Entity)-[:RELATED_TO {relation, weight, source}]->(:Entity)
(:Concept)-[:SUBCLASS_OF]->(:Concept)
(:Document)-[:MENTIONS {position, frequency}]->(:Entity)

7.4.4 金融风控模型

金融交易图:

  (Account)-[:TRANSACTS_TO {amount, time, type}]->(Account)
  (Account)-[:OWNED_BY]->(Person)
  (Person)-[:IDENTIFIED_BY]->(IDDocument)
  (Person)-[:ASSOCIATED_WITH {risk_score}]->(Person)
  (Account)-[:FLAGGED_BY]->(Alert)

7.5 Schema 管理

7.5.1 AgensGraph 的 Schema 特性

AgensGraph 采用灵活 Schema（Schema-optional）模式：

7.5.2 唯一性约束

-- 为 Person.name 创建唯一性约束
CREATE CONSTRAINT person_name_unique
  ON (p:Person)
  ASSERT p.name IS UNIQUE;

-- 为多个属性创建复合唯一约束
CREATE CONSTRAINT person_email_unique
  ON (p:Person)
  ASSERT (p.email) IS UNIQUE;

-- 查看所有约束
-- (在 psql 中执行)
\d person

-- 删除约束
DROP CONSTRAINT person_name_unique;

7.5.3 存在性约束

-- 确保 Person 必须有 name 属性
CREATE CONSTRAINT person_name_exists
  ON (p:Person)
  ASSERT EXISTS(p.name);

-- 确保 KNOWS 关系必须有 since 属性
CREATE CONSTRAINT knows_since_exists
  ON ()-[r:KNOWS]-()
  ASSERT EXISTS(r.since);

7.5.4 约束一览

7.6 模式演化

7.6.1 添加新标签

-- 给现有顶点添加标签
MATCH (p:Person)
WHERE p.role = 'manager'
SET p:Manager
RETURN count(p) AS managers_labeled;

7.6.2 添加新属性

-- 批量添加属性
MATCH (p:Person)
WHERE NOT EXISTS(p.created_at)
SET p.created_at = datetime()
RETURN count(p) AS updated;

-- 修改属性名
MATCH (p:Person)
WHERE EXISTS(p.old_name)
SET p.new_name = p.old_name
REMOVE p.old_name;

7.6.3 数据迁移脚本模板

-- 迁移脚本: v1 → v2
-- 1. 添加新属性
MATCH (p:Person)
SET p.schema_version = 2;

-- 2. 数据转换
MATCH (p:Person)
WHERE p.full_name IS NOT NULL
SET p.first_name = split(p.full_name, ' ')[0],
    p.last_name = split(p.full_name, ' ')[1];

-- 3. 验证
MATCH (p:Person)
WHERE p.schema_version = 2 AND p.first_name IS NULL
RETURN count(p) AS unmigrated;
-- 期望结果: 0

7.7 业务场景：电商平台图建模

7.7.1 需求分析

7.7.2 图模型定义

-- 创建 Schema
-- 顶点
CREATE (:User {user_id, name, email, age, city, registered});
CREATE (:Product {product_id, name, price, category, brand, stock});
CREATE (:Order {order_id, date, total, status, payment_method});
CREATE (:Category {category_id, name, level});
CREATE (:Review {review_id, content, rating, date});

-- 边
(:User)-[:PLACED {at}]->(:Order)
(:Order)-[:CONTAINS {quantity, unit_price}]->(:Product)
(:User)-[:VIEWED {at, duration_sec}]->(:Product)
(:User)-[:FAVORITED {at}]->(:Product)
(:User)-[:WROTE]->(:Review)-[:ABOUT]->(:Product)
(:Product)-[:BELONGS_TO]->(:Category)
(:Category)-[:SUBCATEGORY_OF]->(:Category)
(:Product)-[:SIMILAR_TO {score}]->(:Product)

7.7.3 典型查询

-- 协同过滤推荐：买了此商品的人还买了什么
MATCH (p:Product {name: 'iPhone 15'})<-[:CONTAINS]-(o:Order)-[:PLACED]->(u:User)
MATCH (u)-[:PLACED]->(o2:Order)-[:CONTAINS]->(recommended:Product)
WHERE recommended <> p
RETURN recommended.name, count(*) AS frequency
ORDER BY frequency DESC
LIMIT 10;

-- 用户行为分析
MATCH (u:User {user_id: 'U001'})
OPTIONAL MATCH (u)-[v:VIEWED]->(p:Product)
OPTIONAL MATCH (u)-[f:FAVORITED]->(fp:Product)
OPTIONAL MATCH (u)-[:PLACED]->(o:Order)-[:CONTAINS]->(op:Product)
RETURN u.name,
       count(DISTINCT p) AS viewed_products,
       count(DISTINCT fp) AS favorited_products,
       count(DISTINCT o) AS orders,
       count(DISTINCT op) AS purchased_products;

7.8 建模反模式

7.8.1 超级节点问题详解

超级节点示例: 某明星在社交网络中有 1000 万粉丝

  (明星)-[:FOLLOWS]-(粉丝1)
  (明星)-[:FOLLOWS]-(粉丝2)
  ...
  (明星)-[:FOLLOWS]-(粉丝1000万)

问题:
  - 遍历所有粉丝时性能极差
  - 每次查询都需要处理数百万条边

解决方案:
  1. 边分区: (明星)-[:FOLLOWS_2024]->(粉丝)
  2. 属性过滤: -[:FOLLOWS {tier: 'vip'}]->
  3. 分层: (明星)-[:HAS_FAN_GROUP]->(Group)-[:MEMBER]->(粉丝)

7.9 本章小结

7.10 练习

1. 为"在线教育平台"设计图模型，包括学生、课程、教师、章节、评价等实体。
2. 为你的建模创建完整的约束（唯一性 + 存在性）。
3. 识别以下场景中的超级节点风险并提出改进方案：一个有 500 万用户的电商系统中，热门商品被数万人收藏。

7.11 扩展阅读

• Graph Data Modeling (Neo4j)
• Graph Database Design Patterns
• NoSQL Data Modeling Techniques
• Graph Data Modeling — Practical Guide