Initial commit
This commit is contained in:
@@ -0,0 +1,445 @@
|
||||
# 第二节 图数据建模与Neo4j集成
|
||||
|
||||
> [本节完整代码](https://github.com/datawhalechina/all-in-rag/blob/main/code/C9/rag_modules/graph_data_preparation.py)
|
||||
|
||||
## 一、数据来源与转换
|
||||
|
||||
### 1.1 从Markdown到图数据的转换
|
||||
|
||||
本章的图数据来源于第八章中使用的Markdown格式菜谱数据。为了构建知识图谱,笔者用AI开发了一个简单的[Agent](https://github.com/datawhalechina/all-in-rag/tree/main/code/C9/agent(%E4%BB%A3%E7%A0%81%E7%B3%BBai%E7%94%9F%E6%88%90)),通过LLM将结构化的Markdown菜谱数据转换为CSV格式的图数据。
|
||||
|
||||
**转换流程**:
|
||||
1. **读取Markdown菜谱**:从第八章的数据源加载菜谱文件
|
||||
2. **LLM解析提取**:使用大语言模型识别和提取实体及关系
|
||||
3. **结构化输出**:生成nodes.csv和relationships.csv文件
|
||||
4. **图数据导入**:通过Cypher脚本导入Neo4j数据库
|
||||
|
||||
### 1.2 图数据文件结构
|
||||
|
||||
转换后的图数据包含两个核心文件:
|
||||
|
||||
```
|
||||
data/C9/cypher/
|
||||
├── nodes.csv # 节点数据(菜谱、食材、步骤等)
|
||||
├── relationships.csv # 关系数据(菜谱-食材、菜谱-步骤等)
|
||||
└── neo4j_import.cypher # 数据导入脚本
|
||||
```
|
||||
|
||||
## 二、图数据模型设计
|
||||
|
||||
### 2.1 实际数据结构分析
|
||||
|
||||
基于LLM转换后的实际图数据,知识图谱包含以下核心实体类型。如果你有游戏逆向经验,可以把这些实体类型想象成虚幻引擎烹饪游戏中的对象类,节点间的关系就像对象间的指针引用:
|
||||
|
||||
**核心实体类型**:
|
||||
- **Recipe (菜谱)**:具体的菜品,包含难度、菜系、时间等属性
|
||||
- **Ingredient (食材)**:制作菜品所需的原料,包含分类、用量、单位等
|
||||
- **CookingStep (烹饪步骤)**:详细的制作步骤,包含方法、工具、时间估计
|
||||
- **CookingMethod (烹饪方法)**:如炒、煮、蒸、炸等烹饪技法
|
||||
- **CookingTool (烹饪工具)**:如炒锅、蒸锅、刀具等
|
||||
- **DifficultyLevel (难度等级)**:一星到五星的难度分级
|
||||
- **RecipeCategory (菜谱分类)**:素菜、荤菜、水产、早餐等分类
|
||||
|
||||
**实际数据特点**:
|
||||
- **统一编码体系**:使用nodeId进行唯一标识(如201000001)
|
||||
- **多语言支持**:包含preferredTerm、fsn等多语言字段
|
||||
- **丰富属性**:每个实体包含详细的属性信息
|
||||
- **层次化结构**:从抽象概念到具体实例的层次化组织
|
||||
|
||||
### 2.2 实际节点模型
|
||||
|
||||
基于实际数据的图数据模型:
|
||||
|
||||
```mermaid
|
||||
graph TB
|
||||
%% 定义节点样式
|
||||
classDef recipeNode fill:#e1f5fe,stroke:#01579b,stroke-width:2px
|
||||
classDef ingredientNode fill:#f3e5f5,stroke:#4a148c,stroke-width:2px
|
||||
classDef stepNode fill:#e8f5e8,stroke:#1b5e20,stroke-width:2px
|
||||
classDef categoryNode fill:#fff3e0,stroke:#e65100,stroke-width:2px
|
||||
classDef difficultyNode fill:#fce4ec,stroke:#880e4f,stroke-width:2px
|
||||
|
||||
%% 菜谱节点
|
||||
Recipe["🍽️ Recipe<br/>菜谱节点<br/>---<br/>nodeId: String<br/>name: String<br/>preferredTerm: String<br/>fsn: String<br/>conceptType: String<br/>synonyms: String<br/>category: String<br/>difficulty: Float<br/>cuisineType: String<br/>prepTime: String<br/>cookTime: String<br/>servings: String<br/>tags: String<br/>filePath: String"]
|
||||
|
||||
%% 食材节点
|
||||
Ingredient["🥬 Ingredient<br/>食材节点<br/>---<br/>nodeId: String<br/>name: String<br/>preferredTerm: String<br/>category: String<br/>amount: String<br/>unit: String<br/>isMain: Boolean<br/>synonyms: String"]
|
||||
|
||||
%% 烹饪步骤节点
|
||||
CookingStep["👨🍳 CookingStep<br/>烹饪步骤节点<br/>---<br/>nodeId: String<br/>name: String<br/>description: String<br/>stepNumber: Float<br/>methods: String<br/>tools: String<br/>timeEstimate: String"]
|
||||
|
||||
%% 菜谱分类节点
|
||||
RecipeCategory["📂 RecipeCategory<br/>菜谱分类节点<br/>---<br/>nodeId: String<br/>name: String<br/>preferredTerm: String<br/>fsn: String"]
|
||||
|
||||
%% 难度等级节点
|
||||
DifficultyLevel["⭐ DifficultyLevel<br/>难度等级节点<br/>---<br/>nodeId: String<br/>name: String<br/>preferredTerm: String<br/>fsn: String"]
|
||||
|
||||
%% 关系连接
|
||||
Recipe -->|REQUIRES<br/>需要食材<br/>amount, unit| Ingredient
|
||||
Recipe -->|CONTAINS_STEP<br/>包含步骤<br/>step_order| CookingStep
|
||||
Recipe -->|BELONGS_TO_CATEGORY<br/>属于分类| RecipeCategory
|
||||
Recipe -->|HAS_DIFFICULTY_LEVEL<br/>具有难度| DifficultyLevel
|
||||
|
||||
%% 应用样式
|
||||
class Recipe recipeNode
|
||||
class Ingredient ingredientNode
|
||||
class CookingStep stepNode
|
||||
class RecipeCategory categoryNode
|
||||
class DifficultyLevel difficultyNode
|
||||
```
|
||||
|
||||
**节点类型说明**:
|
||||
|
||||
- **🍽️ Recipe (菜谱节点)**: 核心实体,包含菜谱的完整信息
|
||||
- **🥬 Ingredient (食材节点)**: 制作菜谱所需的食材信息
|
||||
- **👨🍳 CookingStep (烹饪步骤节点)**: 详细的制作步骤和方法
|
||||
- **📂 RecipeCategory (菜谱分类节点)**: 菜品分类(素菜、荤菜、水产等)
|
||||
- **⭐ DifficultyLevel (难度等级节点)**: 制作难度分级(一星到五星)
|
||||
|
||||
### 2.3 实际关系模型
|
||||
|
||||
基于实际数据的关系结构:
|
||||
|
||||
```mermaid
|
||||
graph LR
|
||||
%% 定义节点样式
|
||||
classDef recipeNode fill:#e1f5fe,stroke:#01579b,stroke-width:3px
|
||||
classDef ingredientNode fill:#f3e5f5,stroke:#4a148c,stroke-width:2px
|
||||
classDef stepNode fill:#e8f5e8,stroke:#1b5e20,stroke-width:2px
|
||||
classDef categoryNode fill:#fff3e0,stroke:#e65100,stroke-width:2px
|
||||
classDef difficultyNode fill:#fce4ec,stroke:#880e4f,stroke-width:2px
|
||||
classDef rootNode fill:#f5f5f5,stroke:#424242,stroke-width:2px
|
||||
classDef methodNode fill:#e3f2fd,stroke:#0277bd,stroke-width:2px
|
||||
classDef toolNode fill:#f1f8e9,stroke:#33691e,stroke-width:2px
|
||||
|
||||
%% 核心节点
|
||||
Recipe["🍽️ Recipe<br/>菜谱"]
|
||||
Ingredient["🥬 Ingredient<br/>食材"]
|
||||
CookingStep["👨🍳 CookingStep<br/>烹饪步骤"]
|
||||
RecipeCategory["📂 RecipeCategory<br/>菜谱分类"]
|
||||
DifficultyLevel["⭐ DifficultyLevel<br/>难度等级"]
|
||||
|
||||
%% 层次化节点
|
||||
Root["🌳 Root<br/>根节点"]
|
||||
CookingMethod["🔥 CookingMethod<br/>烹饪方法"]
|
||||
CookingTool["🔧 CookingTool<br/>烹饪工具"]
|
||||
|
||||
%% 主要关系 - 带属性标注
|
||||
Recipe -.->|"REQUIRES<br/>relationshipId: String<br/>amount: String<br/>unit: String<br/><br/>示例: 300g, 2个"| Ingredient
|
||||
Recipe -.->|"CONTAINS_STEP<br/>relationshipId: String<br/>step_order: Float<br/><br/>示例: 1.0, 2.0"| CookingStep
|
||||
Recipe -->|"BELONGS_TO_CATEGORY<br/>菜谱分类关系"| RecipeCategory
|
||||
Recipe -->|"HAS_DIFFICULTY_LEVEL<br/>难度等级关系"| DifficultyLevel
|
||||
|
||||
%% 层次化关系
|
||||
Root -->|"IS_A<br/>概念层次"| Recipe
|
||||
Root -->|"IS_A<br/>概念层次"| Ingredient
|
||||
Root -->|"IS_A<br/>概念层次"| CookingMethod
|
||||
Root -->|"IS_A<br/>概念层次"| CookingTool
|
||||
|
||||
%% 应用样式
|
||||
class Recipe recipeNode
|
||||
class Ingredient ingredientNode
|
||||
class CookingStep stepNode
|
||||
class RecipeCategory categoryNode
|
||||
class DifficultyLevel difficultyNode
|
||||
class Root rootNode
|
||||
class CookingMethod methodNode
|
||||
class CookingTool toolNode
|
||||
```
|
||||
|
||||
**关系类型说明**:
|
||||
|
||||
| 关系编码 | 关系类型 | 说明 | 属性 |
|
||||
|---------|---------|------|------|
|
||||
| **801000001** | REQUIRES | 菜谱-食材关系 | relationshipId, amount, unit |
|
||||
| **801000003** | CONTAINS_STEP | 菜谱-步骤关系 | relationshipId, step_order |
|
||||
| **801000004** | HAS_DIFFICULTY_LEVEL | 菜谱-难度关系 | relationshipId |
|
||||
| **801000005** | BELONGS_TO_CATEGORY | 菜谱-分类关系 | relationshipId |
|
||||
|
||||
**关系特点**:
|
||||
- **虚线箭头**:表示带有丰富属性的关系(如REQUIRES、CONTAINS_STEP)
|
||||
- **实线箭头**:表示简单的分类关系
|
||||
- **层次化结构**:Root节点作为概念层次的顶层节点
|
||||
|
||||
## 三、Neo4j数据导入
|
||||
|
||||
### 3.1 数据准备脚本
|
||||
|
||||
系统通过 `GraphDataPreparationModule` 来处理图数据的加载和管理:
|
||||
|
||||
```python
|
||||
class GraphDataPreparationModule:
|
||||
def __init__(self, neo4j_config: dict):
|
||||
"""
|
||||
初始化图数据准备模块
|
||||
|
||||
Args:
|
||||
neo4j_config: Neo4j连接配置
|
||||
"""
|
||||
self.driver = GraphDatabase.driver(
|
||||
neo4j_config['uri'],
|
||||
auth=(neo4j_config['user'], neo4j_config['password'])
|
||||
)
|
||||
|
||||
def load_graph_data(self) -> List[Dict]:
|
||||
"""
|
||||
从Neo4j加载图数据
|
||||
|
||||
Returns:
|
||||
包含菜谱信息的字典列表
|
||||
"""
|
||||
query = """
|
||||
MATCH (r:Recipe)
|
||||
OPTIONAL MATCH (r)-[:REQUIRES]->(i:Ingredient)
|
||||
OPTIONAL MATCH (r)-[:HAS_STEP]->(s:Step)
|
||||
OPTIONAL MATCH (r)-[:BELONGS_TO]->(c:Category)
|
||||
RETURN r, collect(DISTINCT i) as ingredients,
|
||||
collect(DISTINCT s) as steps,
|
||||
collect(DISTINCT c) as categories
|
||||
ORDER BY r.name
|
||||
"""
|
||||
|
||||
with self.driver.session() as session:
|
||||
result = session.run(query)
|
||||
return [record for record in result]
|
||||
```
|
||||
|
||||
### 3.2 实际CSV数据格式
|
||||
|
||||
转换后的CSV文件格式(基于实际数据):
|
||||
|
||||
**nodes.csv结构**:
|
||||
```csv
|
||||
nodeId,labels,name,preferredTerm,fsn,conceptType,synonyms,category,difficulty,cuisineType,prepTime,cookTime,servings,tags,filePath,amount,unit,isMain,description,stepNumber,methods,tools,timeEstimate
|
||||
```
|
||||
|
||||
**实际数据示例**:
|
||||
```csv
|
||||
201000184,Recipe,干煎阿根廷红虾,干煎阿根廷红虾,,Recipe,"[{'term': '干pan-fried阿根廷红虾', 'language': 'zh'}]",水产,3.0,,提前1天冷藏解冻+10分钟,约5分钟,1人,"趁热吃,柠檬可增酸提味",dishes\aquatic\干煎阿根廷红虾\干煎阿根廷红虾.md,,,,,,,,
|
||||
201000185,Ingredient,阿根廷红虾,阿根廷红虾,,Ingredient,,蛋白质,,,,,,,,2-3,只,True,,,,,
|
||||
201000196,CookingStep,步骤1,步骤1,,CookingStep,,,,,,,,,,,,,阿根廷红虾提前1天从速冻取出放到冷藏里自然解冻,1.0,解冻,冰箱,24小时
|
||||
```
|
||||
|
||||
**relationships.csv结构**:
|
||||
```csv
|
||||
startNodeId,endNodeId,relationshipType,relationshipId,amount,unit,step_order
|
||||
```
|
||||
|
||||
**实际关系示例**:
|
||||
```csv
|
||||
201000184,201000185,801000001,R_000001,2-3,只,
|
||||
201000184,201000196,801000003,R_000010,,,1.0
|
||||
201000184,720000000,801000002,R_000020,,,
|
||||
```
|
||||
|
||||
## 四、图数据查询与检索
|
||||
|
||||
### 4.1 基础查询模式
|
||||
|
||||
#### 简单实体查询
|
||||
```cypher
|
||||
// 查找所有水产类菜谱
|
||||
MATCH (r:Recipe)
|
||||
WHERE r.category = "水产"
|
||||
RETURN r.name, r.difficulty, r.prepTime, r.cookTime
|
||||
|
||||
// 查找包含特定食材的菜谱
|
||||
MATCH (r:Recipe)-[:REQUIRES]->(i:Ingredient)
|
||||
WHERE i.name CONTAINS "虾"
|
||||
RETURN r.name, r.difficulty, i.name, i.amount, i.unit
|
||||
|
||||
// 使用全文搜索查找菜谱
|
||||
CALL db.index.fulltext.queryNodes("recipe_fulltext_index", "川菜 OR 辣椒")
|
||||
YIELD node, score
|
||||
RETURN node.name, node.category, score
|
||||
ORDER BY score DESC
|
||||
```
|
||||
|
||||
#### 多跳关系查询
|
||||
```cypher
|
||||
// 查找某个难度等级的所有菜谱(基于属性查询)
|
||||
MATCH (r:Recipe)
|
||||
WHERE r.difficulty = 3.0
|
||||
RETURN r.name, r.category, r.prepTime, r.cookTime, r.difficulty
|
||||
|
||||
// 查找菜谱的完整制作流程
|
||||
MATCH (r:Recipe {name: "干煎阿根廷红虾"})-[:CONTAINS_STEP]->(s:CookingStep)
|
||||
RETURN r.name, s.stepNumber, s.description, s.methods, s.tools
|
||||
ORDER BY s.stepNumber
|
||||
```
|
||||
|
||||
### 4.2 复杂推理查询
|
||||
|
||||
#### 基于约束的菜谱推荐
|
||||
```cypher
|
||||
// 查找适合新手的简单菜谱(低难度、步骤少)
|
||||
MATCH (r:Recipe)
|
||||
WHERE r.difficulty <= 2.0
|
||||
AND r.stepCount <= 5
|
||||
RETURN r.name, r.difficulty, r.stepCount, r.category
|
||||
ORDER BY r.difficulty, r.stepCount
|
||||
|
||||
// 查找制作时间短的菜谱
|
||||
MATCH (r:Recipe)
|
||||
WHERE r.prepTime IS NOT NULL AND r.cookTime IS NOT NULL
|
||||
AND r.prepTime CONTAINS "分钟" AND r.cookTime CONTAINS "分钟"
|
||||
RETURN r.name, r.prepTime, r.cookTime, r.category
|
||||
ORDER BY r.name
|
||||
```
|
||||
|
||||
#### 菜谱组合推荐
|
||||
```cypher
|
||||
// 查找同一分类下的不同菜谱
|
||||
MATCH (r1:Recipe), (r2:Recipe)
|
||||
WHERE r1.category = r2.category
|
||||
AND r1.category = "水产"
|
||||
AND r1.nodeId <> r2.nodeId
|
||||
RETURN r1.name, r2.name, r1.category
|
||||
LIMIT 5
|
||||
|
||||
// 查找包含相同食材的不同菜谱
|
||||
MATCH (r1:Recipe)-[:REQUIRES]->(i:Ingredient)<-[:REQUIRES]-(r2:Recipe)
|
||||
WHERE r1.nodeId <> r2.nodeId
|
||||
AND i.name = "阿根廷红虾"
|
||||
RETURN r1.name, r2.name, i.name
|
||||
```
|
||||
|
||||
## 五、图数据到文档的转换
|
||||
|
||||
### 5.1 结构化文档构建
|
||||
|
||||
```python
|
||||
def build_recipe_documents(self, graph_data: List[Dict]) -> List[Document]:
|
||||
"""将图数据转换为结构化文档"""
|
||||
|
||||
documents = []
|
||||
for record in graph_data:
|
||||
recipe = record['r']
|
||||
ingredients = record['ingredients']
|
||||
steps = record['steps']
|
||||
categories = record['categories']
|
||||
|
||||
# 构建结构化文档内容
|
||||
content_parts = [
|
||||
f"# {recipe['name']}",
|
||||
f"分类: {', '.join([c['name'] for c in categories])}",
|
||||
f"难度: {recipe['difficulty']}星",
|
||||
# ... 时间、份量等基本信息
|
||||
"",
|
||||
"## 所需食材"
|
||||
]
|
||||
|
||||
# 添加食材列表
|
||||
for i, ingredient in enumerate(ingredients, 1):
|
||||
content_parts.append(f"{i}. {ingredient['name']}")
|
||||
|
||||
content_parts.extend(["", "## 制作步骤"])
|
||||
|
||||
# 添加制作步骤(按顺序排序)
|
||||
sorted_steps = sorted(steps, key=lambda x: x.get('order', 0))
|
||||
for step in sorted_steps:
|
||||
content_parts.extend([
|
||||
f"### 第{step['order']}步",
|
||||
step['description'],
|
||||
""
|
||||
])
|
||||
|
||||
# 创建Document对象
|
||||
document = Document(
|
||||
page_content="\n".join(content_parts),
|
||||
metadata={
|
||||
'recipe_name': recipe['name'],
|
||||
'node_id': recipe.get('nodeId'), # 关键:保持与图节点的关联
|
||||
'difficulty': recipe.get('difficulty', 0),
|
||||
'categories': [c['name'] for c in categories],
|
||||
'ingredients': [i['name'] for i in ingredients]
|
||||
# ... 其他元数据
|
||||
}
|
||||
)
|
||||
documents.append(document)
|
||||
|
||||
return documents
|
||||
```
|
||||
|
||||
> **为什么不直接读取原始Markdown文件?**
|
||||
>
|
||||
> 虽然第八章中HowToCook项目的Markdown格式是统一的,但图RAG的价值在于提供更丰富的信息:
|
||||
>
|
||||
> **原始Markdown的特点**:
|
||||
> - **格式统一**:HowToCook项目有良好的Markdown结构(`#`、`##`、`###`层级)
|
||||
> - **信息完整**:包含菜品名称、原料、制作步骤等基本信息
|
||||
> - **元数据推断**:可以从文件路径推断分类,从`★★★★★`符号推断难度
|
||||
>
|
||||
> **图数据构建文档的额外价值**:
|
||||
> 1. **关系信息丰富**:包含食材间的替代关系、菜谱间的相似性等图关系
|
||||
> 2. **结构化查询**:可以通过图关系快速获取相关信息(如"包含鸡肉的所有菜谱")
|
||||
> 3. **动态内容生成**:根据图关系动态生成推荐内容(如"相似菜谱"、"替代食材")
|
||||
> 4. **语义增强**:图数据库可以存储更丰富的语义信息和计算结果
|
||||
> 5. **查询优化**:图查询在复杂关系检索上比文本搜索更高效
|
||||
|
||||
### 5.2 图RAG中的分块策略
|
||||
|
||||
在图RAG系统中,分块策略与上个项目有所不同,主要体现在**数据来源和上下文获取方式**的差异:
|
||||
|
||||
**图RAG vs 传统RAG的分块对比**:
|
||||
|
||||
| 特性 | 第八章 传统RAG | 第九章 图RAG |
|
||||
|------|-----------------|----------------|
|
||||
| **数据来源** | 直接读取Markdown文件 | 从图数据库构建文档 |
|
||||
| **上下文获取** | 父子文档映射 | 图关系遍历 |
|
||||
| **关系信息** | 有限(仅父子关系) | 丰富(多种图关系) |
|
||||
| **分块策略** | 按Markdown标题分块 | 按语义+长度智能分块 |
|
||||
| **元数据来源** | 文件路径+内容推断 | 图节点结构化数据 |
|
||||
|
||||
**图RAG分块的特点**:
|
||||
1. **保持图关联**:每个chunk通过`parent_id`与图节点关联
|
||||
2. **语义优先分块**:优先按章节分块,保持语义完整性
|
||||
3. **丰富的元数据**:直接从图节点获取结构化信息
|
||||
4. **双重上下文**:既有文本块关系,又有图关系信息
|
||||
|
||||
### 5.3 实际分块实现
|
||||
|
||||
在图RAG系统中,采用的实际分块策略:
|
||||
|
||||
```python
|
||||
def chunk_documents(self, chunk_size: int = 500, chunk_overlap: int = 50) -> List[Document]:
|
||||
"""图RAG文档分块:结合图结构优势的智能分块策略"""
|
||||
|
||||
chunks = []
|
||||
for doc in self.documents:
|
||||
content = doc.page_content
|
||||
|
||||
if len(content) <= chunk_size:
|
||||
# 短文档:保持完整,避免破坏语义
|
||||
chunk = Document(
|
||||
page_content=content,
|
||||
metadata={
|
||||
**doc.metadata,
|
||||
"parent_id": doc.metadata["node_id"], # 关键:保持与图节点的关联
|
||||
"chunk_index": 0,
|
||||
"doc_type": "chunk"
|
||||
}
|
||||
)
|
||||
chunks.append(chunk)
|
||||
else:
|
||||
# 长文档:智能分块策略
|
||||
sections = content.split('\n## ')
|
||||
|
||||
if len(sections) <= 1:
|
||||
# 无章节结构:按长度分块(带重叠)
|
||||
total_chunks = (len(content) - 1) // (chunk_size - chunk_overlap) + 1
|
||||
for i in range(total_chunks):
|
||||
start = i * (chunk_size - chunk_overlap)
|
||||
end = min(start + chunk_size, len(content))
|
||||
# ... 创建chunk,保持parent_id关联
|
||||
else:
|
||||
# 有章节结构:按语义分块(推荐)
|
||||
for i, section in enumerate(sections):
|
||||
chunk_content = section if i == 0 else f"## {section}"
|
||||
# ... 创建chunk,包含section_title信息
|
||||
|
||||
return chunks
|
||||
```
|
||||
|
||||
图RAG的分块策略在保持语义完整性的基础上,充分利用图数据库的结构化优势。与第八章直接读取Markdown文件不同,这里从图数据库构建标准化文档,每个chunk通过`parent_id`与原始Recipe节点保持关联,既继承了传统的父子文档映射关系,又能通过图关系遍历获取更丰富的上下文信息。在具体实现上,采用智能分块策略:短文档保持完整避免破坏语义,长文档优先按`##`标题进行章节分块,必要时才进行长度分块,同时为每个chunk提供丰富的元数据(如chunk_id、chunk_index、total_chunks等),确保后续处理的灵活性和可追溯性。
|
||||
|
||||
Reference in New Issue
Block a user