Amazon Comprehend

• Amazon Comprehend
  • Overview
  • Why Use Amazon Comprehend?
  • Custom Classification
  • Named Entity Recognition (NER)
  • Custom Entity Recognition
  • Amazon Comprehend Medical

Overview

• A fully managed, serverless natural language processing (NLP) service
• It accepts input documents which can come from a wide area of sources such as: social media, email, web pages, raw documents, transcripts, medical records (Comprehend Medical)
• Uses machine learning to find insights and relationships within text
• Key Capabilities:
  • Extracts key phrases, entities (people, places, brands, events)
  • Analyzes sentiment (positive or negative)
  • Understands language and performs tokenization
  • Part-of-speech tagging
  • Organizes collections of text files by topics
  • Events detection (mainly about companies)
  • Identifies and redacts PII (Personally Identifiable Information)
  • Targeted sentiment analysis for specific entities
  • Custom classification and entity recognition
• We can train it on our own data
• Use Cases
  • Customer Interaction Analysis
    • Analyze emails and support tickets to identify factors leading to positive or negative experiences
    • Categorize customer feedback by sentiment and topics
    • Identify common complaint patterns
    • Content Organization
      • Group articles by topics that Comprehend automatically uncovers
      • Organize large document collections by theme
      • Discover topics without manual categorization

Example Workflow:

Source

Why Use Amazon Comprehend?

• Fully Managed: No infrastructure to maintain
• Serverless: Scales automatically
• Pre-trained Models: Ready to use out-of-the-box
• Customizable: Train on your own data for specific use cases
• Multiple Analysis Types: Real-time and asynchronous processing

Custom Classification

• We can organize documents into categories that we define
• Example: categorize customer emails so that we can provide guidance based on the type of the customer request
• Supports different types of documents (txt, PDF, Word, images)
• Real-time analysis on single documents on a synchronous way
• Asynchronous analysis on a batch of documents

Named Entity Recognition (NER)

• NER – Extracts predefined, general-purpose entities like people, places, organizations, dates, and other standard categories, from text
• Use Cases
  • Identify key people in a document
  • Extract company names and locations
  • Find dates and times
  • Categorize events and activities

Custom Entity Recognition

• We can analyze text for specific terms and noun-based phrases
• Can be used to extract terms like policy numbers or phrases that imply a customer escalation, or anything specific to our business
• We can train the model with custom data such as a list of entities and documents that contain them
• It can do real-time or async analysis

Amazon Comprehend Medical

• Detects and returns useful information in unstructured clinical text:
  • Can understand physician’s notes
  • Can discharge summaries
  • Can understand test results and case notes
• Uses NLP to detect Protected Health Information (PHI)
• We can store input documents in Amazon S3
• Input data can be analyzed in real-time with the help of Kinesis Data Firehose
• We can use Amazon Transcribe to transcribe patient narratives into text that can be analyzed by Comprehend Medical

Example Workflow:

Input Text:

Output Entities:

Source