# 知识图谱问答

> 2024年后，更多参考[graphRAG设计](/mle-interview/03_system/03_ml/rag.md)

## 1. requirements

**Functional Requirements:**

* Natural language question input
* Accurate answers based on knowledge graph information
* Support for complex multi-hop reasoning
* Handle different question types (factoid, relationship, comparison)

**Non-Functional Requirements:**

* Low latency (<500ms response time)
* High availability (99.9%)
* Scalability to handle large knowledge graphs
* Support multiple languages
* Privacy and security compliance

## 2. ML task & pipeline

![](/files/O0fhllDLdeNBjzaeH8vl)

![](/files/yAtJC4GMp2yXAYIgO9wO)

a) Question Understanding:

* Question type classification
* Entity recognition and linking
* Relation extraction
* Query intent classification

b) Graph Processing:

* Graph embedding generation
* Subgraph retrieval
* Path ranking

c) Answer Generation:

* Answer extraction/generation
* Confidence scoring
* Evidence compilation

## 3. data collection

**Sources:**

* Knowledge Graphs:
  * Wikidata
  * DBpedia
  * Domain-specific KGs
  * Company internal KGs
* Training Data:
  * WebQuestions
  * ComplexWebQuestions
  * LC-QuAD 2.0
  * KQA Pro
  * MetaQA

**Data Processing:**

* Entity normalization
* Relation alignment
* Graph completion
* Question-answer pair generation

## 4. feature

**Question Features:**

* BERT/RoBERTa embeddings
* Dependency parsing features
* Named entity mentions
* Question type indicators

**Graph Features:**

* Node embeddings (TransE, RotatE)
* Structural features
* Path features
* Subgraph features

## 5. model

## 6. evaluation

**Metrics:**

* Accuracy
* F1 Score
* Hits\@K
* MRR (Mean Reciprocal Rank)
* Path validity
* Answer completeness
* Reasoning correctness

**Testing Approaches:**

* Unit tests for each component
* Integration tests
* A/B testing
* Human evaluation
* Adversarial testing

## 7. deployment & serving

Infrastructure:

* Containerization with Docker
* Kubernetes for orchestration
* GPU support for inference
* Load balancing
* Auto-scaling

## 8. monitor & maintenance

Monitoring:

* Model performance metrics
* System health metrics
* Error rates and types
* Latency distribution
* Resource utilization

## reference

* <https://github.com/shawnh2/QA-CivilAviationKG>
* [美团知识图谱问答技术实践与探索](https://tech.meituan.com/2021/11/03/knowledge-based-question-answering-in-meituan.html)
* [检索式对话系统在美团客服场景的探索与实践](https://tech.meituan.com/2022/11/03/retrieval-based-dialogue-system.html)
* [阿里小蜜：知识结构化推动智能客服升级](https://mp.weixin.qq.com/s/x9CkAyLKgLj7E7K1F2Q6iA)
* [阿里实时语音与智能对话](https://mp.weixin.qq.com/s/scvTTqApSr8SbCKRlUoz-g)
* [智能机器人在滴滴出行场景的技术探索](https://mp.weixin.qq.com/s/MSy8OHzR3avObmOq9uSSFQ)
* [QQ浏览器智能问答技术探索实践](https://mp.weixin.qq.com/s/nN0aSXQN_IyjIJ1mRT5s3w)
* [达摩院基于元学习的对话系统](https://mp.weixin.qq.com/s/Ji_-hTe5vwpnyu-whj3PXg)
* [大话知识图谱-意图识别和槽位填充](https://zhuanlan.zhihu.com/p/165963264)


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://longxingtan.gitbook.io/mle-interview/03_system/03_ml/knowledge_graph_qa.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.