Comprehend

🧠 What is Amazon Comprehend?

Amazon Comprehend is a fully managed NLP service that uses machine learning to extract insights and relationships from text — no ML expertise required.

✅ It can detect sentiment, key phrases, named entities, language, and even supports custom classification and entity recognition.

---

📦 Key Use Cases

Use Case	Description
🧠 Sentiment analysis	Understand tone in product reviews, emails, tweets, etc.
🏷️ Entity extraction	Extract names, dates, organizations, locations
🔍 Text classification	Categorize support tickets, feedback, news, etc.
📑 Key phrase extraction	Pull important terms from paragraphs
🌍 Language detection	Identify the language of the given text
🧾 Custom classifiers	Train models to classify content in domain-specific ways
📄 Custom entity recognition	Detect domain-specific entities (e.g., product codes, IDs)
🏥 Medical text analysis	(Comprehend Medical) Extract protected health info (PHI)

---

🚀 Features Overview

Feature	API / Feature Name	Description
Language detection	DetectDominantLanguage	Auto-detects the language of the input text
Entity recognition	DetectEntities	Extracts people, places, dates, organizations
Key phrase detection	DetectKeyPhrases	Highlights important phrases
Sentiment analysis	DetectSentiment	Positive, Negative, Neutral, Mixed
Syntax analysis	DetectSyntax	Parts of speech tagging (noun, verb, etc.)
PII detection	ContainsPiiEntities	Finds PII (email, SSN, phone, etc.)
Custom classification	CreateDocumentClassifier	Train and use a classifier with your labeled dataset
Custom entity recognition	CreateEntityRecognizer	Train model to detect domain-specific named entities
Comprehend Medical	comprehendmedical:*	Specialized medical text analysis (HIPAA compliant)

---

🌐 Supported Languages (Standard Comprehend)

• English, Spanish, French, German, Italian, Portuguese

• Hindi (limited), Japanese, Korean, Chinese (Simplified)

• Many APIs (like sentiment) support 6–10 languages

---

💡 Example: Analyze a Product Review

Input:

“The new headphones are amazing! Great sound and battery life.”

Output:

• Sentiment: Positive

• Entities: headphones (Product)

• Key Phrases: new headphones, great sound, battery life

---

🧪 Python (Boto3) Examples

🔍 Detect Sentiment

import boto3

client = boto3.client('comprehend')

response = client.detect_sentiment(
    Text="I love this product. It works flawlessly.",
    LanguageCode='en'
)

print(response['Sentiment'])  # POSITIVE

---

📌 Detect Key Phrases

response = client.detect_key_phrases(
    Text="The movie had great visual effects and soundtrack.",
    LanguageCode='en'
)

for phrase in response['KeyPhrases']:
    print(phrase['Text'], " - Confidence:", phrase['Score'])

---

🏷️ Detect Entities

response = client.detect_entities(
    Text="Barack Obama was the 44th President of the United States.",
    LanguageCode='en'
)

for entity in response['Entities']:
    print(entity['Type'], entity['Text'])

---

🧑‍🏫 Custom Classification (Advanced)

You can:

• Train a custom classifier using labeled text documents in S3 (CSV format)

• Create and train with CreateDocumentClassifier

• Use ClassifyDocument API on new text

Dataset Format (CSV)

__label__billing   The customer has a billing issue with their last invoice.
__label__support   I can't connect to the internet since yesterday.

---

💸 Pricing (2024)

Feature	Price (per unit)
Text analysis (standard)	~$1.00 per 1000 units (100 units = 100 tokens ≈ 100 words)
Custom classification	~$3.00 per hour (training) + usage fees
PII detection	~$0.0001 per unit
Comprehend Medical	~$0.0015 per unit
Free tier	50K units/month for 12 months

---

🛡️ Security & Compliance

Feature	Support
IAM control	✅ Yes
Logging with CloudTrail	✅ Yes
HIPAA compliance	✅ Comprehend Medical
KMS integration	✅ For S3 training data
VPC endpoints	✅ Via AWS PrivateLink

---

🧱 Terraform Support

Comprehend support in Terraform is limited to custom classifier resources.

Custom Classifier Example:

resource "aws_comprehend_document_classifier" "my_classifier" {
  document_classifier_name = "support-ticket-classifier"
  data_access_role_arn     = aws_iam_role.comprehend_access.arn
  language_code            = "en"
  input_data_config {
    s3_uri     = "s3://my-dataset/classifier/train.csv"
    data_format = "COMPREHEND_CSV"
  }
}

---

🔗 Service Integrations

AWS Service	Integration Use Case
S3	Store input documents or training datasets
Lambda	Analyze text as it is uploaded or submitted
SNS / SQS	Alert when classification is complete
Translate	Translate → Comprehend for multi-language support
QuickSight	Analyze sentiment trends or classification breakdowns
OpenSearch	Index structured text from Comprehend analysis

---

✅ TL;DR Summary

Feature	Amazon Comprehend
Ready-made NLP	✅ Yes
Language detection	✅ Yes
Sentiment & key phrases	✅ Yes
Entity recognition	✅ Yes (standard and custom)
PII detection	✅ Yes
Custom classification	✅ Yes (via S3 CSV)
Terraform support	⚠️ Partial (custom only)
Free tier	✅ 50K units/mo for 12 months
Common integration	Lambda, Translate, S3, QuickSight

---