Semantic Natural Language Understanding with Spark, UIMA & Machine Learned Ontologies
How Hollywood Learned to Love the Semantic Web
-
Upload
chris-testa -
Category
Technology
-
view
30.180 -
download
0
description
Transcript of How Hollywood Learned to Love the Semantic Web
![Page 1: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/1.jpg)
Lauren Conrad 1.7 Million Twitter Followers
HOW HOLLYWOOD LEARNED TO LOVE THE SEMANTIC WEB
![Page 2: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/2.jpg)
Mandy Moore 2.3 Million Twitter Followers
A SEEMINGLY SIMPLE PROCESS
• We match brands with celebrities that reach their target audiences on Twitter.
• We write endorsement copy that is authentic to the celeb and will resonate with followers.
• Through OAuth to celeb’s Twitter accounts, we distribute approved endorsements.
• We count the conversations the
campaign inspires, plus clicks-though, retweets, etc.
![Page 3: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/3.jpg)
WHY CELEBRITIES?
Kim Kardashian 7.2 Million Twitter Followers
• Top Celebrities can get 15,000 clicks per tweet.
• The NYT and WSJ average about 400.
• Celebrities cut through the noise.
![Page 4: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/4.jpg)
BBC AMERICA’S TOP GEAR
![Page 5: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/5.jpg)
BBC AMERICA’S TOP GEAR
We count the Clicks
We count the Retweets
We count the Conversations
![Page 6: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/6.jpg)
AUTHENTIC CONVERSATIONS
BBC campaign encouraged people to share stories of their first car. The hashtag #myfirstcar was distributed by celebrities who seeded the conversation with stories of their first vehicle. Success Metric: created over 15,000 responses for #myfirstcar, highest-rated show in Top Gear history
#myfirstcar #myfirstcar
#myfirstcar
![Page 7: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/7.jpg)
TOYOTA SIENNA – SWAGGER WAGON
![Page 8: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/8.jpg)
TOYOTA SIENNA – SWAGGER WAGON
We count the Clicks
We count the Video Views
We count the Retweets
2.8 MM Views
![Page 9: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/9.jpg)
DEVELOPING A UNIQUE EXPERTISE:
Ad.ly CONNECTS BRANDS WITH CONSUMERS VIA TOP CELEBS
Cristiano Ronaldo 2.5 Million Twitter Followers
• 1,000 of the top celebrities, artists and athletes on Twitter.
• 24,000 successful celebrity endorsements in 18 mos.
• 150 Brands: NBC, Sony, Best Buy, Old Navy, Microsoft, etc.
![Page 10: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/10.jpg)
DEVELOPING A UNIQUE EXPERTISE:
Ad.ly CONNECTS BRANDS WITH CONSUMERS VIA TOP CELEBS
Cristiano Ronaldo 2.5 Million Twitter Followers
• 1,000 of the top celebrities, artists and athletes on Twitter.
• 24,000 successful celebrity endorsements in 18 mos.
• 150 Brands: NBC, Sony, Best Buy, Old Navy, Microsoft, etc.
• 100K Prospects: accounts on Twitter over 10K followers
![Page 11: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/11.jpg)
Paul Pierce 1.8 Million Followers
• Needed to get our arms around celebrity data & make it actionable.
• Must integrate with all aspects of the
business from pre-to-post-campaign.
CHALLENGE: CODIFY OUR TRIBAL KNOWLEDGE
0 20000 40000 60000 80000 100000 120000
Jan-‐10
Jun-‐10
Nov-‐10
Apr-‐11
Sep-‐11
Feb-‐12
# Celebs
![Page 12: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/12.jpg)
HOW CAN WE DO THIS?
![Page 13: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/13.jpg)
My Profile Studied at Maryland Did a project @ MINDSWAP with Prof. Jim Hendler on early SPARQL implementation Did the IBM Extreme Blue Internship Program working on OLAP Went to Google / YouTube – not much culture of SemTech Never deployed semantic technologies until… Ad.ly: • Had a pressing business need • Had limited resources • Had limited funds
Disclaimer: I was a skep0c
![Page 14: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/14.jpg)
FREEBASE SATIATED MY SKEPTIC
http://www.freebase.com/view/en/kim_kardashian
/en/kim_kardashian
![Page 15: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/15.jpg)
WHY LINKED DATA IS WORKING
Tera-sized datasets are now available for real world concepts. Celebrities are well-annotated. Lots of industry interest: • Google buys Freebase, dataset grows • Linked Data explodes • Facebook’s OpenGraph embeds RDFa
everywhere • Facebook’s Graph API brings graph datasets to
the masses w/ REST+JSON
Great data & technology licensing terms for business. • Freebase Acre is a great way to prototype This just wasn’t true 6 years ago.
![Page 16: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/16.jpg)
How: 5 steps to integrating linked data
UNDERSTAND WHAT YOUR “THINGS” ARE
CHOOSE A LINKED DATASET
RECONCILE YOUR THINGS
1
2
3
BUILD BUSINESS INTELLIGENCE
FEEDBACK & MAINTENANCE
4
5
![Page 17: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/17.jpg)
ADLY DATA MODEL
Endorsement* Adver@ser* Celebrity*
150 Total 24K Total 1K Total
Expected 10K in 6 months Prospect 100K inside of a year
*All entities have performance & analytics data
![Page 18: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/18.jpg)
ADLY CELEBRITY
Snoop Dogg (Celebrity)
hHp://twiHer.com/snoopdogg
hHp://facebook.com/snoopdogg
Followers: 3.3M
Gender: 54.3% Male, 45.7% Female
Top Ci@es: LA, NYC, Chicago, Atlanta, DC
Top Countries: USA, India, Philippines
Avg RTs per Ad: 21
Fans: 8.9M
…
Avatar: aaa
![Page 19: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/19.jpg)
How: 5 steps to integrating linked data
UNDERSTAND WHAT YOUR “THINGS” ARE
CHOOSE A LINKED DATASET
RECONCILE YOUR THINGS
1
2
3
BUILD BUSINESS INTELLIGENCE
FEEDBACK & MAINTENANCE
4
5
![Page 20: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/20.jpg)
CHOOSING WHAT TO LINK TO
Linking Open Data cloud diagram, by Richard Cyganiak and Anja Jentzsch. http://lod-cloud.net/
![Page 21: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/21.jpg)
TOUR OF LINKED DATA
![Page 22: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/22.jpg)
How: 5 steps to integrating linked data
UNDERSTAND WHAT YOUR “THINGS” ARE
CHOOSE A LINKED DATASET
RECONCILE YOUR THINGS
1
2
3
BUILD BUSINESS INTELLIGENCE
FEEDBACK & MAINTENANCE
4
5
![Page 23: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/23.jpg)
ADLY CELEBRITY
Snoop Dogg
hHp://twiHer.com/snoopdogg
hHp://facebook.com/snoopdogg
Followers: 3.3M
Gender: 54.3% Male, 45.7% Female
Top Ci@es: LA, NYC, Chicago, Atlanta, DC
Top Countries: USA, India, Philippines
Avg RTs per Ad: 21
Fans: 8.9M
…
Avatar: aaa
![Page 24: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/24.jpg)
ADLY CELEBRITY WITH FREEBASE
Snoop Dogg
hHp://twiHer.com/snoopdogg
hHp://facebook.com/snoopdogg
Followers: 3.3M
Gender: 54.3% Male, 45.7% Female
Top Ci@es: LA, NYC, Chicago, Atlanta, DC
Top Countries: USA, India, Philippines
Avg RTs per Ad: 21
Fans: 8.9M
…
Avatar: aaa
hHp://freebase.com/en/snoop_dogg
Image: aaa
Aliases: Calvin Cordozar Broadus, Jr.
Professions: Rapper, Musician, Actor
Date of Birth: 10-‐20-‐1971
Gender: Male
Marital Status: Married
![Page 25: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/25.jpg)
RECONCILING WITH FREEBASE
• As new Celebrities come into our system, they are added to the “Match Queue”
• When 2 experts agree on a match, the Celebrity is “reconciled” with the freebase entity
New Celebrity enters queue
Judgement s@ll out
Skip
New
Freebase ID Matched!
2x Confirmation
![Page 26: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/26.jpg)
RECONCILING WITH FREEBASE
![Page 27: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/27.jpg)
RECONCILING WITH FREEBASE
MATCH!
![Page 28: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/28.jpg)
RECONCILING WITH FREEBASE
![Page 29: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/29.jpg)
RECONCILING WITH FREEBASE No clear match
![Page 30: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/30.jpg)
ADLY DATA WITH FREEBASE
Snoop Dogg
hHp://twiHer.com/snoopdogg
hHp://facebook.com/snoopdogg
Followers: 3.3M
Gender: 54.3% Male, 45.7% Female
Top Ci@es: LA, NYC, Chicago, Atlanta, DC
Top Countries: USA, India, Philippines
Avg RTs per Ad: 21
Fans: 8.9M
…
Avatar: aaa
hHp://freebase.com/en/snoop_dogg
Image: aaa
Aliases: Calvin Cordozar Broadus, Jr.
Professions: Rapper, Musician, Actor
Date of Birth: 10-‐20-‐1971
Gender: Male
Marital Status: Married
![Page 31: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/31.jpg)
FREEBASE RELATIONAL MAPPING
• Special shout out to author Jeff Schenck • Also could be called FreebaseAlchemy • RDFAlchemy already exists!
![Page 32: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/32.jpg)
DATA FLOW
<code> c.freebase_date_of_birth
</code> Freebase
MQL Query
web1 cache1
<code> c.name </code>
db1 SQL Query
web1
DATABASE PROPERTY:
FREEBASE PROPERTY:
![Page 33: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/33.jpg)
FREEBASE TOOLS
http://matchmaker.freebaseapps.com/
Matchmaker
http://freebase.com/queryeditor
Schema Explorer http://schemas.freebaseapps.com/
Acre http://acre.freebase.com/
MQL Query Editor
![Page 34: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/34.jpg)
How: 5 steps to integrating linked data
UNDERSTAND WHAT YOUR “THINGS” ARE
CHOOSE A LINKED DATASET
RECONCILE YOUR THINGS
1
2
3
BUILD BUSINESS INTELLIGENCE
FEEDBACK & MAINTENANCE
4
5
![Page 35: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/35.jpg)
DEMO: FILTERING
![Page 36: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/36.jpg)
DEMO: AUGMENT VISUAL DISPLAY
https://admin.ad.ly/admin/celebrity/5391/bio/
![Page 37: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/37.jpg)
PLAYING WITH SEMANTIC DATA
WHAT YOU CAN DO • Provide context on reconciled
entities
• Filter on Properties – We do this in our Celeb Finder
• Machine learning on Properties – Add into clustering as signal
• Reasoning
– Need to serialize internal facts into RDF & create Ontology
– Query a SPARQL endpoint
TOOLS I’D USE TO DO IT
Cwm
![Page 38: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/38.jpg)
How: 5 steps to integrating linked data
UNDERSTAND WHAT YOUR “THINGS” ARE
CHOOSE A LINKED DATASET
RECONCILE YOUR THINGS
1
2
3
BUILD BUSINESS INTELLIGENCE
FEEDBACK & MAINTENANCE
4
5
![Page 39: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/39.jpg)
FEEDBACK & MAINTENANCE
Tell us what’s b0rked!
Negative Feedback
Op-Amp Abstraction
In Out
![Page 40: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/40.jpg)
FEEDBACK QUEUE
![Page 41: How Hollywood Learned to Love the Semantic Web](https://reader033.fdocuments.us/reader033/viewer/2022042813/54577956b1af9fb66e8b52e2/html5/thumbnails/41.jpg)
Summary
• You don’t need a cross PhD in Computer Science and Philosophy to use SemTech today for visual augmentation and business model properties
• Linking datasets benefit from human intervention, but tools like the Reconciliation Queue makes this task easier & scalable
• This 5 step process sets you up to do long term advanced Semantic Analysis with Reasoning, Machine Learning, and so much more