Machine Learning
Amazon Comprehend
Amazon Comprehend applies managed natural-language processing to text for entities, sentiment, language, key phrases, PII, and supported custom models. It can add deterministic preprocessing and classification around a generative model.
Key points
- PII detection returns entity types and character offsets
- Asynchronous jobs process documents stored in S3
- Custom classification and entity recognition require trained resources
When to use it
- Redact detected PII before prompt logging
- Route incoming text by language or business category before model selection
Exam tips
- Choose Comprehend for text NLP; Macie discovers sensitive objects in S3 and Rekognition analyzes images
- Offsets and confidence values require application handling, especially with transformed Unicode text