情感分析

1. requirements

constraint

• latency: how long it takes a single request

• throughput: how many request can be handled in a given amount of time

2. ML task & pipeline

3. data collection

• 收集data

  • GDPR（privacy），数据脱敏，数据加密

• 分析data。考虑label的distribution

• 考虑feature是不是只有text的，还是有numeric，nominal的。missing data怎么处理

4. feature

• text的feature怎么生成embedding，好处坏处有哪些。（word embedding, fasttext, BERT）

• numeric的missing data,如何normalize

• 实际工作中，都是每个ML组都有自己不同的embedding set。互相使用别人的embedding set。怎么pre-train, fine-train, 怎么combine feature

5. model

• 模型选择: 传统模型还是神经网络

• 考虑系统方面的constraint, 如prediction latency, memory. 怎么合理的牺牲模型的性能以换取constraint方面的benefit

• 模型蒸馏

6. evaluation

• train, test, validation split data

• evaluation matrix

• feature的ABtest怎么做

7. deploy & serving

• GPU or CPU

• 单机多进程 or Spark + Broadcast, KF-serving

• dynamic batching

• Dynamic Model Input (输入数据的长度)

• quantization (cast)

• distill/or smaller model

• onnx

  • 不同的硬件和推理引擎兼容

  • 进一步优化: 算子融合、内存优化和硬件加速

• caching responses to reduce the request

8. Monitor & maintenance

• hardware usage

• serving usage: qps

• model performance

• business object

9. 优化与问答

• train/test data和product上distribution不一样怎么办

• data distribution 随着时间改变怎么办

reference

• 细粒度情感分析在到餐场景中的应用
• 情感分析技术在美团的探索与应用
• learn.microsoft.com/en-us/azure/ai-services
• Using Sentiment Score to Assess Customer Service Quality
• System Design of Extreme Multi-label Query Classification using a Hybrid Model
• Query理解在美团搜索中的应用 - DataFunTalk的文章 - 知乎
• How to Fine-Tune BERT for Text Classification?
• How We Scaled Bert To Serve 1+ Billion Daily Requests on CPUs
• FastFormers: Highly Efficient Transformer Models for Natural Language Understanding
• Understanding Pins through keyword extraction
• 华为云细粒度文本情感分析及应用
• ML System Design Mock Interview - Build an ML System That Classifies Which Tweets Are Toxic