Amazon Comprehend
Natural-language processing: entities, sentiment, PII, topics.
❓ What is it?
A prebuilt NLP service that extracts insight from text — entities, key phrases, sentiment, language, syntax, topics — and detects or redacts PII, with custom classification and custom entity recognition when the built-ins are not enough.
💡 Why does it exist?
Most organisations sit on mountains of unstructured text (reviews, tickets, contracts) they cannot query. Comprehend turns that text into structured signals via an API call, with zero model training for the common cases.
⏱️ When should you use it?
Use it for text analytics: routing support tickets by topic, mining review sentiment, redacting PII from documents before storage, or classifying documents with a small labeled set (custom classifier).
🗺️ Where does it fit?
Downstream of text producers: S3 batch jobs for archives, synchronous API calls inside a Lambda for live text, and as the NLP stage in document pipelines after Textract extracts the raw text.
🔌 How do you integrate it?
Call DetectSentiment / DetectEntities / DetectPiiEntities synchronously, or start async analysis jobs over S3 prefixes; train custom classifiers/entity recognizers by supplying labeled examples in S3.
🧩 Commonly integrated with
🎯 Exam angle (AIF-C01)
- Text + "sentiment / entities / PII detection" → Comprehend. Medical text has a dedicated variant (Comprehend Medical).
- Textract reads documents into text; Comprehend understands the text — pipelines often chain them in that order.