the united states postal service processed over 150 billion pieces of mail in 2013—far too much...

Post on 21-Dec-2015

215 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Spark the future.

May 4 – 8, 2015Chicago, IL

Azure Machine Learning MarketplaceElad ZiklikGroup ManagerAzure ML Marketplace

BRK2560

Who is this guy ?

Big question for FIFA SOCCER 2014 World Cup :

“ Who will be Paul II ?”

….and the winner is

“ Accurate predictions in 15 out of 16 games in K.O. stage

Cortana did not stop !

Oscars!

What is this graph?

Some #HowOldNumbers

In 4 days: • 33M users• 1.2M visitors /

h• 236M Images• 10K images /

sec• 1.7M FB

Shares

The United States Postal Service processed over 150 billion pieces of mail in 2013—far too much for efficient human sorting.

But as recently as 1997, only 10% of hand-addressed mail was successfully sorted automatically.

ML in the “real world”

The challenge in automation is enabling computers to interpret endless variation in handwriting.

The challenge in automation is enabling computers to interpret endless variation in handwriting.

By providing feedback, the Postal Service was able to train computers to accurately read human handwriting.

Today, with the help of machine learning, over 98% of all mail is successfully processed by machines.

Machine Learning - using past data to predict the future

Recommenda-tion engines

Advertising analysis

Weather forecasting for business planning

Social network analysis

IT infrastructure and web app optimization

Legal discovery and document archiving

Pricing analysisFraud detection

Churn analysis

Equipment monitoring

Location-based tracking and services

Personalized Insurance

Imagine what Machine Learning could do to your

business

IntroducingZiklikMart

Classical market basket analysis Increases cross-sell and up-sell, provides pricing opportunities Most retailers use manual business rules to offer upsells, cross

sells

Appears on almost every Amazon item page

Frequently bought together

Make machine learning accessible to every enterprise, data scientist, developer, information worker, consumer, and device anywhere in the world.

Azure Machine Learning Vision

Make machine learning accessible to every enterprise, data scientist, developer, information worker, consumer, and device anywhere in the world.

Azure ML Vision - Marketplace

ML APIs

Marketplace

ML Operationalization

ML Studio

ML Algorithms

Azure ML Marketplace

ProducerConsumer

JJ Foods & ML Frequently Bought Together

Customers Quick service restaurants Retailing Educatering Pubs and hotels Event catering

Year 2014 8 sites 60,000 customers 4,500 unique products 1 million sales orders £185 million turnover

JJ Food Service

2 recommendation /personalization scenarios Item-specific recommendations Checkout-specific

recommendations

6% of items added to cart come from Azure ML models

5% Conversion rate at checkout

JJ Food Service

Product Recommendati

ons

Basket Analysis

Customer Churn

Prediction

Text Analytics

Anomaly Detection

Azure Machine Learning APIs

Customer ChurnPrediction

Churn Interactive Experience

For every business, some customer churn – leave

What if we can tell ahead of time which customers are going to leave?

Machine Learning Algorithm can “learn” from previous history transaction patterns

Use the trained model to make predictions

Target customers most likely to leave

Customer Churn Prediction – How?

JJ Foods & Customer Churn Prediction

Customers Quick service restaurants Retailing Educatering Pubs and hotels Event catering

Year 2014 8 sites 60,000 customers 4,500 unique products 1 million sales orders £185 million turnover

JJ Food Service

Top 20% of churn predictions are over 3x better than random pick

Out of 200 predicted using our system ~45% retained Past experience without our

system was ~17% retention rate

JJ Food Service

Text Analytics ServiceAnalyzes unstructured text.Product reviews, support tickets, emails, etc.

Sentiment AnalysisHow do your customers feel about your brand or products?

Key Phrase ExtractionWhat are your customers talking about?

Sentiment analysis

Key phrase extraction

“It was a wonderful hotel, with unique décor and friendly staff”

“It was a wonderful hotel, with unique décor and friendly staff”Interactive Experience

Text Analytics Service Use Cases

Sentiment of Brand/Product

Customer Support Triage

User ReviewsTopic Extraction

Jul-13 Aug-13

Sep-13

Oct-13

Nov-13

Dec-13

Jan-14 Feb-14

Mar-14

Apr-14

May-14

Jun-14 Jul-14 Aug-14

Sep-14

Oct-14

Nov-14

Dec-14

Jan-15 Feb-15

0.65

0.66

0.67

0.68

0.69

0.7

0.71

0.72

0.73

Sentiment

Sentim

ent

Overall Sentiment over Time

Jul-13 Aug-13

Sep-13

Oct-13

Nov-13

Dec-13

Jan-14 Feb-14

Mar-14

Apr-14

May-14

Jun-14 Jul-14 Aug-14

Sep-14

Oct-14

Nov-14

Dec-14

Jan-15 Feb-15

0.65

0.66

0.67

0.68

0.69

0.7

0.71

0.72

0.73

Sentiment

Sentim

ent

Overall Sentiment over Time

Oct 2013

Oct 2014

Top Key Phrases with Positive Sentiment

Negative21%

Neutral15%

Positive64%

Negative

Neutral

Positive

ScorePositive sentiment: above 0.65Negative sentiment: below 0.4Neutral sentiment: between 0.4 & 0.65

% of Users by Sentiment

Fans

UserAvg Sentiment

Score# of Text Pieces

user1723 0.390934 8

user228 0.323264 6

user1250 0.392574 6

user955 0.245724 5

user229 0.28559 5

user518 0.328739 5

user1532 0.398888 5

user4963 0.179529 4

user2541 0.212446 4

user312 0.221297 4

UserAvg Sentiment

Score# of Text Pieces

user77 0.758117 69

user23 0.833552 58

user243 0.716238 53

user396 0.974862 40

user1403 0.716291 38

user215 0.929712 35

user862 0.838113 27

user40 0.972512 27

user780 0.730805 26

user268 0.746828 22

Critics

Potential Evangelists Fix your relationship

Anomaly Detection

SLA 99.99%

Behind the scenes of the #HowOldRobot – Introducing Project Oxford

A portfolio of REST APIs and SDKs which enable developers to write applications which understand the content within the rapidly growing set of multimedia data

Project Oxford’s API services will help you understand and interact with audio, text, image, and video

Understand the data around your application

Microsoft Project Oxford Services

PROJECT OXFORD

Speech APIs LUIS (Language Understanding Intelligent Service)

Vision APIs Face APIs

OCR

Speech Recognition

Text to Speech

Speech Intent Recognition

Determine Entities

Behind #HowOldRobot - Microsoft Project Oxford

PROJECT OXFORD

Face Grouping

Face Identification

Face DetectionAnalyze Image

Generate Thumbnail

Improve Models

Detect Intent

Face APIsVision APIsLUIS(Language Understanding Intelligent Service)

Speech APIs

Face APIsDetectionVerificationGroupingIdentification

Face API – DetectionDetection Result:JSON:[ { "faceRectangle": { "width": 109, "height": 109, "left": 62, "top": 62 }, "attributes": { "age": 31, "gender": "male", "headPose": { "roll": "2.9", "yaw": "-1.3", "pitch": "0.0" } "faceLandmarks": { "pupilLeft": { "x": "93.6", "y": "88.2" }, "pupilRight": { "x": "138.4", "y": "91.7" }, ...

INPUTIMAGE

FACIALRECTANGLE + LANDMARKS

DETECTION ATTRIBUTES

Verification Result:JSON:[

{"isIdentical":false,"confidence":0.01

}]

Face API – Verification

Given two faces, determine whether they are the same person

CLUSTERED BY DETECTED PEOPLE

Face API – Grouping

He is Chao Wang.

Face API – Identify

NEW INPUTIMAGE

IDENTIFY

Natalie Huber

GROUP PERSON OBJECTS

COLLEAGUES

RECOGNITION

Chao Wang

Vision APIsAnalyze an ImageOCRGet Thumbnail

Understand content and features within an image

Analyze Image Service

Analyze Image – Example

Type of Image:

Clip Art Type 0 Non-clipart

Line Drawing Type 0 Non-Line Drawing

Black & White Image False

Content of Image:

Categories [{ “name”: “people_swimming”, “score”: 0.099609375 }]

Adult Content False

Adult Score 0.18533889949321747

Faces [{ “age”: 27, “gender”: “Male”, “faceRectangle”: {“left”: 472, “top”: 258, “width”: 199, “height”: 199}}]

Image Colors:

Dominant Color Background White

Dominant Color Foreground Grey

Dominant Colors White

Accent Color

Detect and recognize words within a photo

OCR Service

OCR – Example

JSON:{

"language": "en","orientation": "Up","regions": [

{

"boundingBox": "41,77,918,440","lines": [

{

"boundingBox": "41,77,723,89",

"words": [

{

"boundingBox": "41,102,225,64",

"text": "LIFE"

},

{

"boundingBox": "356,89,94,62",

"text": "IS"

},

{

"boundingBox": "539,77,225,64",

"text": "LIKE"

}. . .

TEXT:LIFE IS LIKERIDING A BICYCLETO KEEP YOUR BALANCEYOU MUST KEEP MOVING

Stay tuned to our Azure ML blog https://aka.ms/mlblog

Publishers welcome ! Be a part of this new data science economy. Contact us at MLAppsSupport@microsoft.com

Check out the Machine Learning API gallery at gallery.azureml.net

Discover right now the Azure Machine Learning APIs ready to boost your services

Get started at azure.com/ml

Ignite Azure Challenge Sweepstakes

Attend Azure sessions and activities, track your progress online, win raffle tickets for great prizes!

Aka.ms/MyAzureChallenge

Enter this session code online: “NGFC”

(10) - Microsoft Surface Pro 3Core i5 256GB

(30) – Xbox One Master Chief Collection Bundle

(55) – Microsoft Band

NO PURCHASE NECESSARY. Open only to event attendees. Winners must be present to win. Game ends May 9th, 2015. For Official Rules, see The Cloud and Enterprise Lounge or myignite.com/challenge

Offers throughout the week

Ignite Azure Challenge Sweepstakes

Attend Azure sessions and activities, track your progress online, win raffle tickets for great prizes!

Aka.ms/MyAzureChallenge

Enter this session code online: BRK2560

NO PURCHASE NECESSARY. Open only to event attendees. Winners must be present to win. Game ends May 9th, 2015. For Official Rules, see The Cloud and Enterprise Lounge or myignite.com/challenge

Visit Myignite at http://myignite.microsoft.com or download and use the Ignite Mobile App with the QR code above.

Please evaluate this sessionYour feedback is important to us!

© 2015 Microsoft Corporation. All rights reserved.

top related