knowledge-synthesizer

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Knowledge Synthesizer

Purpose

目的

Provides expertise in aggregating information from multiple sources and synthesizing it into structured, actionable knowledge. Specializes in ontology building, knowledge graph design, and insight extraction for RAG and AI systems.

擅长将多来源信息进行聚合，并合成为结构化、可落地的知识。专注于本体构建、知识图谱设计，以及为RAG和AI系统提取洞察。

When to Use

适用场景

Building knowledge graphs or ontologies
Designing GraphRAG or hybrid retrieval systems
Synthesizing information across multiple documents
Extracting entities and relationships from text
Creating structured knowledge bases
Developing taxonomy and classification systems
Implementing semantic search architectures
Connecting disparate data sources meaningfully

构建知识图谱或本体
设计GraphRAG或混合检索系统
跨多份文档合成信息
从文本中提取实体与关系
创建结构化知识库
开发分类法与分类系统
实现语义搜索架构
有意义地连接分散的数据源

Quick Start

快速开始

Invoke this skill when:

Building knowledge graphs or ontologies
Designing RAG systems with graph components
Synthesizing insights from multiple sources
Extracting structured knowledge from unstructured text
Creating taxonomies or classification schemes

Do NOT invoke when:

Vector database setup without graph needs → use
```
/context-manager
```
General NLP tasks (NER, classification) → use
```
/nlp-engineer
```
Database schema design → use
```
/database-administrator
```
Document writing → use
```
/technical-writer
```

在以下场景调用此技能：

构建知识图谱或本体
设计包含图组件的RAG系统
从多来源合成洞察
从非结构化文本中提取结构化知识
创建分类法或分类方案

请勿在以下场景调用：

无需图功能的向量数据库设置 → 使用
```
/context-manager
```
通用NLP任务（命名实体识别、分类）→ 使用
```
/nlp-engineer
```
数据库架构设计 → 使用
```
/database-administrator
```
文档撰写 → 使用
```
/technical-writer
```

Decision Framework

决策框架

Knowledge Structure Needed?
├── Hierarchical (taxonomy)
│   └── Tree structure, parent-child relationships
├── Graph (connected entities)
│   └── Nodes + edges, property graphs
├── Hybrid (RAG + Graph)
│   └── Vector embeddings + knowledge graph
└── Flat (simple retrieval)
    └── Standard vector store sufficient

是否需要知识结构化？
├── 层级式（分类法）
│   └── 树形结构、父子关系
├── 图结构（关联实体）
│   └── 节点 + 边、属性图
├── 混合式（RAG + 图）
│   └── 向量嵌入 + 知识图谱
└── 扁平化（简单检索）
    └── 标准向量存储即可满足需求

Core Workflows

核心工作流

1. Ontology Design

1. 本体设计

Identify domain scope and boundaries
Define core entity types (classes)
Map relationships between entities
Add properties and constraints
Validate with domain experts
Document with examples

确定领域范围与边界
定义核心实体类型（类）
映射实体间的关系
添加属性与约束
与领域专家验证
附带示例进行文档记录

2. Knowledge Graph Construction

2. 知识图谱构建

Extract entities from source documents
Identify relationships between entities
Normalize and deduplicate entities
Build graph structure (nodes, edges)
Add metadata and provenance
Create query interfaces

从源文档中提取实体
识别实体间的关系
实体规范化与去重
构建图结构（节点、边）
添加元数据与来源信息
创建查询接口

3. Insight Synthesis

3. 洞察合成

Gather sources and establish provenance
Extract key claims and facts
Identify contradictions and agreements
Synthesize into coherent narrative
Cite sources for traceability
Highlight confidence levels

收集来源并记录信息出处
提取关键主张与事实
识别矛盾与共识
合成为连贯的叙述
标注来源以实现可追溯性
标注置信度等级

Best Practices

最佳实践

Maintain provenance for all extracted knowledge
Use established ontology standards (OWL, SKOS) when applicable
Design for evolution—ontologies change over time
Validate extracted relationships with source context
Balance granularity with usability
Include confidence scores for extracted facts

为所有提取的知识保留来源记录
适用时使用已确立的本体标准（OWL、SKOS）
为演进而设计——本体随时间变化
结合源上下文验证提取的关系
在粒度与易用性间取得平衡
为提取的事实添加置信度评分

Anti-Patterns

反模式

Anti-Pattern	Problem	Correct Approach
No provenance tracking	Cannot verify claims	Track source for every fact
Over-complex ontology	Hard to maintain and query	Start simple, evolve as needed
Ignoring contradictions	Inconsistent knowledge base	Flag and resolve conflicts
Static schema	Breaks with new domains	Design for extensibility
Blind extraction trust	Hallucinated relationships	Validate with confidence thresholds

反模式	问题	正确做法
不跟踪来源	无法验证主张	为每个事实跟踪来源
本体过于复杂	难以维护与查询	从简单开始，按需演进
忽略矛盾	知识库不一致	标记并解决冲突
静态架构	新增领域时失效	为可扩展性而设计
盲目信任提取结果	生成幻觉关系	结合置信度阈值进行验证