Curriculum Overview: Entity Recognition with Azure AI Language
Identify features and uses for entity recognition
Curriculum Overview: Entity Recognition with Azure AI Language
This curriculum provides a structured pathway to mastering Named Entity Recognition (NER) as part of the Microsoft Azure AI Fundamentals (AI-900) certification. It focuses on identifying, categorizing, and applying entity recognition features within Natural Language Processing (NLP) workloads.
Prerequisites
Before beginning this module, learners should have a foundational understanding of the following:
- Basic AI Concepts: Knowledge of common AI workloads (Unit 1), specifically the difference between Computer Vision and NLP.
- General NLP Principles: Understanding that computers process text by breaking it into tokens (words/sentences).
- Azure Fundamentals: Familiarity with the Azure Portal and the general purpose of Azure AI services.
- Cloud Literacy: A basic grasp of SaaS/PaaS models (knowing that Azure provides pre-built models via APIs).
Module Breakdown
| Module | Topic | Difficulty | Focus Area |
|---|---|---|---|
| 1 | Fundamentals of NER | Beginner | Definition, terminology, and core "Big Idea." |
| 2 | The NER Workflow | Intermediate | Preprocessing, algorithms, and classification steps. |
| 3 | Specialized Detection | Intermediate | PII (Personal) and PHI (Health) data redaction. |
| 4 | NER vs. Entity Linking | Advanced | Distinguishing identification from knowledge-base mapping. |
| 5 | Implementation & Scenarios | Practical | Real-world use cases in legal, finance, and search. |
Learning Objectives per Module
Module 1: Fundamentals of NER
- Define Named Entity Recognition as the process of identifying meaningful and distinct parts of information (people, places, dates) in unstructured text.
- Understand how NER tags essential information automatically to facilitate data analysis.
Module 2: The NER Workflow
- Describe the internal pipeline of an NER model.
- Visual Anchor: The NER Pipeline
Module 3: Specialized Detection (PII & PHI)
- Identify the features of Personally Identifiable Information (PII) and Protected Health Information (PHI) detection.
- Understand the use of redaction for data privacy compliance.
Module 4: NER vs. Entity Linking
- Distinguish between Entity Recognition (identifying a "person") and Entity Linking (connecting that person to a specific Wikipedia entry or database record).
- Comparison Matrix:
| Feature | Named Entity Recognition (NER) | Entity Linking |
|---|---|---|
| Goal | Categorize tokens into types. | Connect tokens to a knowledge base. |
| Output | Label (e.g., "Location") | URL/ID (e.g., "Wikipedia: New York") |
| Example | Finds "Venus" as a Planet. | Identifies "Venus" specifically as the Second Planet. |
Module 5: Real-World Application
- Analyze how NER enhances search engines by providing context to keywords.
- Identify how chatbots use NER to extract user details like names or account numbers for personalization.
Success Metrics
To demonstrate mastery of this curriculum, the learner must be able to:
- Identify correct services: Given a scenario (e.g., "Extracting contract dates"), select Azure AI Language as the correct tool.
- Differentiate Features: Correctly choose between NER, Sentiment Analysis, and Key Phrase Extraction in multiple-choice exams.
- Explain the 'Why': Articulate how NER reduces manual sorting and human error in data-heavy sectors like legal and finance.
- Score Achievement: Achieve at least 80% on practice assessments for AI-900 Unit 4.
Real-World Application
Entity recognition is not just a theoretical concept; it is the backbone of modern data-driven automation.
[!IMPORTANT] The "Human Review" Reducer: In the legal and financial sectors, NER can process thousands of documents in seconds, extracting client names and financial figures that would otherwise require weeks of manual human review.
Visualizing Entity Classification:
- Customer Service: Chatbots recognize a user's problem type and name to provide a bespoke experience.
- Search Relevance: Search engines move beyond simple keyword matching to understanding that "Paris" in a query refers to a Location, not a person.