#GlobalAzure
#GlobalAzure
May 14th, 2022
#GlobalAzure
#GlobalAzure
#GlobalAzure
#Bulgaria
#GlobalAzure
• Software Architect @
o 20+ years professional experience
• Microsoft Azure MVP
• External Expert Eurostars-Eureka, Horizon Europe
• External Expert InnoFund Denmark, RIF Cyprus
• Business Interests
o Web Development, SOA, Integration
o IoT, Machine Learning, Computer Intelligence
o Security & Performance Optimization
• Contact
ivelin.andreev@icb.bg
www.linkedin.com/in/ivelin
www.slideshare.net/ivoandreev
#GlobalAzure
• What’s New in Cognitive Services
• Language Studio Services
• Advanced and Custom Modeling
• Setting up Language Studio
• Demo
#GlobalAzure
#GlobalAzure
#Bulgaria
Azure Cognitive for
Language (Overview)
#GlobalAzure
LUIS.ai
Text Analytics
QnA Maker
Entity Recognition
#GlobalAzure
First to achieve human parity
Globally available in multiple languages
Deployment Options
Common Scenarios
https://docs.microsoft.com/en-us/connectors/cognitiveservicestextanalytics/
#GlobalAzure
https://language.cognitive.azure.com/home
Set of UI-based tools
Features
Azure TextAnalytics Resource
#GlobalAzure
Feature Description Studio REST/SDK Docker
Named Entity
Recognition (NER)
Identifies entities in text across several pre-defined
categories
Custom NER AI model to extract custom entity categories, using
unstructured text
Language detection Evaluates text, and determines the language it was written
Entity Linking Disambiguates the identity of an entity found in text and
provides links on Wikipedia
Personally Identifiable
Information (PII)
Identifies entities in text across several pre-defined
categories of sensitive information
Text analytics for
health
Extracts information from unstructured medical texts,
such as clinical notes and doctor's notes.
Sentiment analysis &
Opinion mining
Sentiment labels (negative, neutral, positive) on granular
level for the attributes of a product or service
#GlobalAzure
Feature Description Studio REST/SDK Docker
Key phrase extraction Evaluates unstructured text for list of key phrases and main
points
Text summarization Feature extracts key sentences that collectively convey the
essence of a document. Generate summary.
Custom text
classification
Build an AI model to classify unstructured text into custom
classes that you define
Conversational lang.
understanding
AI model to bring the ability to understand natural
language into apps, bots, and IoT devices
Question answering Provides answers to questions extracted from text input,
using semi-structured content such as: FAQs, manuals, and
documents
Orchestration
workflow
Train language models to connect to multiple services:
question answering, conversational language understanding,
and LUIS
#GlobalAzure
Prices apply to both Container and API
Billing Unit
Tracking AZ Portal Text Analytics resource
Free (5000 text/month)
Pay-as-you-Go (Azure Standard)
Commitment tiers
#GlobalAzure
#GlobalAzure
#GlobalAzure
#Bulgaria
Predefined Studio Services
#GlobalAzure
Combines 4 outputs in a single call
1. Named Entity Recognition
2. Relation Extraction
3. Entity Linking of Medical Concepts
4. Assertion Detection
#GlobalAzure
#GlobalAzure
Extension of classic sentiment analysis (no extra cost)
Higher granularity of target and assessment
Gain insights from the sentiment
Very suitable to call for action
#GlobalAzure
Extraction
Text Summarization
#GlobalAzure
Improved model quality
Personally Identifiable Information (PII)
Custom NER
#GlobalAzure
multiple operations in a single request
Same AI model
Up to 125K characters
throughput, not latency
partial feature support
#GlobalAzure
Feature Limitation
Preconfigured Features • Max request size: 1MB
• Max 25 docs (Async API), 25 (Text summarization) 10 docs (Sentiment, Key
Phrase, Opinion), 5 docs (NER, PII)
• S Tier-1000 req/sec, S0-Tier 100 req/sec per feature
Custom Text Classification • .txt only, no empty files
• West US 2, West Europe regions
• 20 req/minute, Max request size 125’000 chars
• Min 10, Recommended 50 tagged instances per class
• Min 10 files, Max 1’000’000 files, Max 10 models per project
Custom NER • (Same as Custom Text)
• Max file size is 128’000 chars
• Min 10 files, max 100’000 files, Max 50 models per project
Conversational Language
Understanding
• 15’000 utterances (500 chars), 500 intents, 100 entities
• 60 req/min
• Max 100’000 req/month
QnA • File types: .tsv (10MB), .pdf (25MB), .txt (10MB), .docx (10MB), .xlsx (3MB)
• Max question 1’000 chars, Max answer 25’000 chars
• Max QnAs per call: 1000
#GlobalAzure
#GlobalAzure
#Bulgaria
Setting up Language Studio
#GlobalAzure
https://language.cognitive.azure.com/
Register Resource
Provider
#GlobalAzure
#GlobalAzure
https://www.microsoft.com/en-us/ai/responsible-ai
Do not use; Carefully consider; Avoid
#GlobalAzure
#GlobalAzure
#Bulgaria
Advanced & Custom
#GlobalAzure
Test
Deploy
View
Performance
Train
Model
Tag
Utterances
Build
Schema
Custom model to predict meaning and get information
Scenarios
#GlobalAzure
NLP for conversational apps over data
Scenarios
Workflow
#GlobalAzure
Build AI models to classify text into custom categories
Scenarios
Workflow
#GlobalAzure
Custom AI model for domain specific entities in text
Scenarios
Workflow
#GlobalAzure
KPIs
FP
FN
Confusion Matrix
#GlobalAzure
5-step Learning Tutorial (from 4000+ guides)
• https://docs.microsoft.com/en-us/learn/paths/explore-natural-language-processing/
Language Studio Setup
https://docs.microsoft.com/en-us/azure/cognitive-services/language-service/language-studio
Try it out
https://language.cognitive.azure.com/tryout/sentiment
https://language.cognitive.azure.com/tryout/healthAnalysis
https://docs.microsoft.com/en-us/azure/cognitive-services/language-service/custom-
classification/quickstart?pivots=language-studio
Form Recognizer (Preview)
• https://formrecognizer.appliedai.azure.com/studio
• https://docs.microsoft.com/azure/applied-ai-services/form-recognizer
#GlobalAzure
#GlobalAzure
#Bulgaria
DEMO
#GlobalAzure

Language Studio and Custom Models