Comprehend

Amazon Comprehend NLP service for extracting insights from text using natural language processing.

Configuration

Property	Value
Protocol	AwsJson1_1
Signing Name	comprehend
Target Prefix	Comprehend_20171127
Persistence	No

Quick Start

Detect sentiment and entities in a piece of text:

# Detect sentiment
curl -s http://localhost:4566 \
  -H "Content-Type: application/x-amz-json-1.1" \
  -H "X-Amz-Target: Comprehend_20171127.DetectSentiment" \
  -H "Authorization: AWS4-HMAC-SHA256 Credential=test/20260421/us-east-1/comprehend/aws4_request, SignedHeaders=host, Signature=fake" \
  -d '{"Text":"I absolutely love using AWSim for local development! It saves so much time.","LanguageCode":"en"}'

# Detect entities
curl -s http://localhost:4566 \
  -H "Content-Type: application/x-amz-json-1.1" \
  -H "X-Amz-Target: Comprehend_20171127.DetectEntities" \
  -H "Authorization: AWS4-HMAC-SHA256 Credential=test/20260421/us-east-1/comprehend/aws4_request, SignedHeaders=host, Signature=fake" \
  -d '{"Text":"Jeff Bezos founded Amazon in Seattle in 1994.","LanguageCode":"en"}'

Operations

• DetectEntities — identify named entities in text

  • Input: Text (required), LanguageCode (e.g., en, es, fr, de, it, pt, ar, hi, ja, ko, zh)
  • Returns: Entities list, each with Text, Type (PERSON, LOCATION, ORGANIZATION, DATE, QUANTITY, EVENT, TITLE, OTHER), Score (confidence 0.0–1.0), BeginOffset, EndOffset

• DetectKeyPhrases — extract key phrases from a block of text

  • Input: Text, LanguageCode
  • Returns: KeyPhrases list, each with Text, Score, BeginOffset, EndOffset

• DetectSentiment — determine the overall sentiment of text

  • Input: Text, LanguageCode
  • Returns: Sentiment (POSITIVE, NEGATIVE, NEUTRAL, MIXED), SentimentScore with Positive, Negative, Neutral, Mixed float values summing to 1.0

• DetectDominantLanguage — detect the dominant language of a document

  • Input: Text (no LanguageCode needed)
  • Returns: Languages list, each with LanguageCode (BCP-47 code, e.g., en) and Score (confidence)

• BatchDetectEntities — run entity detection on multiple documents in one call

  • Input: TextList (list of strings, max 25), LanguageCode
  • Returns: ResultList (list of entity results by index), ErrorList (any failed items)

• BatchDetectKeyPhrases — run key phrase extraction on multiple documents

  • Input: TextList (max 25), LanguageCode
  • Returns: ResultList, ErrorList

Curl Examples

# 1. Detect key phrases
curl -s http://localhost:4566 \
  -H "Content-Type: application/x-amz-json-1.1" \
  -H "X-Amz-Target: Comprehend_20171127.DetectKeyPhrases" \
  -H "Authorization: AWS4-HMAC-SHA256 Credential=test/20260421/us-east-1/comprehend/aws4_request, SignedHeaders=host, Signature=fake" \
  -d '{"Text":"The quick brown fox jumps over the lazy dog near the river bank.","LanguageCode":"en"}'

# 2. Detect dominant language
curl -s http://localhost:4566 \
  -H "Content-Type: application/x-amz-json-1.1" \
  -H "X-Amz-Target: Comprehend_20171127.DetectDominantLanguage" \
  -H "Authorization: AWS4-HMAC-SHA256 Credential=test/20260421/us-east-1/comprehend/aws4_request, SignedHeaders=host, Signature=fake" \
  -d '{"Text":"Bonjour le monde, comment allez-vous?"}'

# 3. Batch entity detection across multiple documents
curl -s http://localhost:4566 \
  -H "Content-Type: application/x-amz-json-1.1" \
  -H "X-Amz-Target: Comprehend_20171127.BatchDetectEntities" \
  -H "Authorization: AWS4-HMAC-SHA256 Credential=test/20260421/us-east-1/comprehend/aws4_request, SignedHeaders=host, Signature=fake" \
  -d '{"TextList":["Apple was founded by Steve Jobs.","Microsoft is headquartered in Redmond, Washington."],"LanguageCode":"en"}'

SDK Example

import {
  ComprehendClient,
  DetectSentimentCommand,
  DetectEntitiesCommand,
  DetectKeyPhrasesCommand,
  DetectDominantLanguageCommand,
} from '@aws-sdk/client-comprehend';

const comprehend = new ComprehendClient({
  region: 'us-east-1',
  endpoint: 'http://localhost:4566',
  credentials: { accessKeyId: 'test', secretAccessKey: 'test' },
});

const text = 'I love using AWS services in Seattle! The pricing is great.';

// Detect sentiment
const { Sentiment, SentimentScore } = await comprehend.send(new DetectSentimentCommand({
  Text: text,
  LanguageCode: 'en',
}));
console.log('Sentiment:', Sentiment); // POSITIVE
console.log('Scores:', SentimentScore);

// Detect entities
const { Entities } = await comprehend.send(new DetectEntitiesCommand({
  Text: text,
  LanguageCode: 'en',
}));
console.log('Entities:', Entities?.map(e => `${e.Text} (${e.Type})`));
// e.g., ["Seattle (LOCATION)", "AWS (ORGANIZATION)"]

// Detect language
const { Languages } = await comprehend.send(new DetectDominantLanguageCommand({
  Text: text,
}));
console.log('Language:', Languages?.[0]?.LanguageCode); // en

Behavior Notes

• AWSim's Comprehend uses heuristic algorithms rather than real ML models:
  • Entity detection uses pattern matching for common types: capitalized words become PERSON or ORGANIZATION, dates match DATE, numbers match QUANTITY.
  • Sentiment analysis uses positive/negative word lists — texts with words like "love", "great", "excellent" score POSITIVE.
  • Language detection uses simple n-gram character frequency analysis to pick the closest language.
  • Key phrase extraction identifies noun phrase patterns using basic POS tagging.
• Results are approximate and intended for integration testing of your NLP pipeline, not production NLP use.
• All operations are fully stateless — no data is retained between calls.
• Confidence scores (Score) are simulated and do not reflect real model confidence.