Ian Cameron, VP Media & Entertainment
AI
MAKE THE HIDDEN SEEN
Presentation Prepared for
DIGIPUBLISH
Agenda
≠
AI
• Expert System is the largest European
vendor of Artificial Intelligence & Text
Analytics software and solutions
• Public company (EXSY) with offices in
Europe and North America; R&D labs in
Italy, France, Spain and the USA
• 250+ global employees
• Award-winning, patented technology
• The Cognitive Computing technology of
choice for enterprises in all sectors and
government agencies
Global Organization
Selected Media & Entertainment References
Who is Ian Cameron?
Artificial Intelligence:
What is it?
 Sometimes it seems like science fiction
(or magic), but Artificial Intelligence is
always scientific and predictable
 As a science, it advances step by step,
with continuous improvements: a
sudden revolution from one day to the
next is highly improbable
 AI matches the results of some human
cognitive processes using approaches
and methods that are different from
the ones of the human brain
More importantly is it product or a strategy?
• Natural Language Processing (NLP)
• Natural Language Generation (NLG)
• Robotic Process Automation (RPA)
• Machine Learning
• Supervised
• Unsupervised
• Voice to Text (V2T)
• Image Recognition
• Deep Learning
• BioMetrics
• ????
What is AI?
IDEA #1 –Today External Content enters your content storage system
Wires/Feeds
• Financial
• Subscription (AP,
Reuters, etc.)
Archive/Legacy
Content
Social
Media
Public
Domain
Current
Content
CMS/DAM
An AI strategy in your workflow tomorrow
AI Strategy for
Content Enrichment
Content classification
i.e. Identification of people names,
places, events, organizations, topics,
facts and quotes in content
Related stories and
content recommendation
Potential Information Workflow
Wires/Feeds
Archival
& Current
Client CMS/DAM
Bureau/Newsroom
News
Sports
Fashion &
Entertainment
Sentiment
Emotion
Analytics
Trailers
Reviews
Analytics
Enriched Content in your CMS/DAM
Advertising/Marketing Film/Television
Social
Media
Public
Domain
Automated Source
Collection & Alerting
Content Creation
Classification Methodology
AI enhanced network connects knowledge
to content creators
ROI – Idea #1
AI Enriched Content – Upstream
• Increased Content Generation = Incremental Revenue Increase
• Automated Categorization & Extraction = Reduced Production Costs
• Asset Harmonization = Discoverability
• Disambiguated Knowledge Graph = Relevant Insights
Idea #2 - Use an AI strategy to automate Ontology/Taxonomy
creation. (classification strategy)
Some definitions:
Ontology : a conceptual model representing a specific field of information
• Classes (e.g. “person”, “organization”, “dog”)
• Attributes of classes (“age”, “birth date”, “foundation date”)
• Relationships between classes (“marry”, “own”, “hire”)
Taxonomy : is the science of classification according to a predetermined system – hierarchical
• Taxonomy is the process of naming and classifying things such as animals and plants into groups within a
larger system, according to their similarities and differences.
Knowledge Base (KB) : contains the ontology model plus the instances of
• Classes (“John Smith”, “Expert System”, “Buddy”),
• Properties (“John Smith is 25 years old”)
• Relationships (“Expert System hired John Smith”)
• Adding instances to an ontology model = “populating” the ontology
High-Level Workflow
METADATA
APPLICATIONS
SEARCH
ANALYTICS
VISUALIZATION
FACETS
RECOMMENDATIONS
PORTALS
TOPIC PAGES
LINKED DATA
KNOWLEDGE BASES
ONTOLOGY
DESIGN
DATABASE/BIDOCUMENTS
NLP
ENGINE
NLP
DESIGN
…
Subject
Matter ExpertSME
End-User
U
CREATE
ONTOLOGY
MODEL
1.
NLP ENGINE
FEEDS YOUR
ONTOLOGY
3.
CREATE
NLP
PIPELINE
2.
ROI – Idea #2
• Feeding your ontology/taxonomy is essential
• AI strategy leverages semantic reasoning to :
• Automatically feed your ontology/taxonomy based on your content
• Resolve ambiguity for a comprehensive view of each entity
• Maintain up-to date knowledge
• The result ? A well-fed ontology/taxonomy
• Provides end-users with relevant insights based upon knowledge extracted
from your content
Copyright © 2018 Expert System - All Rights Reserved - Slide 17 Confidential – do not distribute
Deep indexing and Data Extraction at
• Goals : Dow Jones Professional Information Business
• Drive a Compelling User Experience with Efficient Search & Alerting
• Create a Differentiated Data Offering
• The Solution : Expert System
• Speed and Quality critical
• Combining Classification and Extraction approaches
• 11 languages
• Classification
• Applying Dow Jones Intelligent Identifiers
• 3000+ Region/Topic/Industries
• Extraction
• First Area : Regulatory & Compliance :
• Special Interest People & Adverse Media Entities
Processing 1.5 Million Articles Every Day
‘‘ AI is critical
to our business’’
Bob Pashinsky
Director, Content Strategy & Metadata
Copyright © 2018 Expert System - All Rights Reserved - Slide 19 Confidential – do not distribute
Creating Meaningful Metadata at
• Goals
• Tag movie clips with meaningful metadata
• Resolve cost and inconsistency of Manual indexing
• The Solution : Expert System AI Capable Deployment
• Harmonize terminology and clip tagging based on SONY term taxonomy
• Real-time Taxonomy validation & enhancement
• Indexers spend less time validating tags
• Seamless integration with other platforms via Expert System REST API’s
Overview
AI : State of the Art
 AI is here to stay: there are many
problems that can be solved right now
in a very effective way
 To implement a solution based on AI,
you need experts, time, analysis and
hard work it is no different than
anything else in the world
 Some use cases which may seem to be
a good fit for AI are not, cost analysis
before implementation of any AI
strategy may show that it is not cost
effective
AI after the Hype
 It is not the solution to all world’s
problems and not the most important
after the discovery of the wheel
 It is not so advanced that a small
amount of work can solve complex
problems and it cannot learn without
supervision;
 Remember what Edison said about
genius:
“Genius is one percent inspiration and
ninety-nine percent perspiration”
AI after the Hype
 It is not machine learning or deep
learning: ML is only a very common
strategy used in many AI use cases
 It is not an omniscient program that
only needs to be configured or trained
to solve real problems
 It is not something that programs itself
and functions without the help of
experts which anyone can easily adapt
to solve real problems
AI : State of the Art
 Implementing an AI solution always
means programming a computer, in
one way or another
 In many cases, the best way to
understand if an AI project can be a
success is to do a Proof of Concept
 AI is not the General Intelligence of
human beings
 An existing process doesn’t need to be
necessarily replicated in the same way
when AI is used
Hints for a Successful AI
Strategy
 A good amount of healthy skepticism is
always a good start to set the right
expectations for AI
 Have confidence in the value of AI and
of experience: sometimes the road can
be long but at the end the results are
real, solid and measurable
 Start with a gradual approach: targeting
a step-by-step implementation of an AI
Strategy is often the best recipe for
success
Thank You / Your
Questions
Ian Cameron, VP Media & Entertainment
icameron@expertsystem.com
443 979 3869

Ian Cameron

  • 1.
    Ian Cameron, VPMedia & Entertainment AI MAKE THE HIDDEN SEEN Presentation Prepared for DIGIPUBLISH
  • 2.
  • 3.
    • Expert Systemis the largest European vendor of Artificial Intelligence & Text Analytics software and solutions • Public company (EXSY) with offices in Europe and North America; R&D labs in Italy, France, Spain and the USA • 250+ global employees • Award-winning, patented technology • The Cognitive Computing technology of choice for enterprises in all sectors and government agencies Global Organization
  • 4.
    Selected Media &Entertainment References
  • 5.
    Who is IanCameron?
  • 6.
    Artificial Intelligence: What isit?  Sometimes it seems like science fiction (or magic), but Artificial Intelligence is always scientific and predictable  As a science, it advances step by step, with continuous improvements: a sudden revolution from one day to the next is highly improbable  AI matches the results of some human cognitive processes using approaches and methods that are different from the ones of the human brain
  • 7.
    More importantly isit product or a strategy? • Natural Language Processing (NLP) • Natural Language Generation (NLG) • Robotic Process Automation (RPA) • Machine Learning • Supervised • Unsupervised • Voice to Text (V2T) • Image Recognition • Deep Learning • BioMetrics • ???? What is AI?
  • 8.
    IDEA #1 –TodayExternal Content enters your content storage system Wires/Feeds • Financial • Subscription (AP, Reuters, etc.) Archive/Legacy Content Social Media Public Domain Current Content CMS/DAM
  • 9.
    An AI strategyin your workflow tomorrow AI Strategy for Content Enrichment Content classification i.e. Identification of people names, places, events, organizations, topics, facts and quotes in content Related stories and content recommendation
  • 10.
    Potential Information Workflow Wires/Feeds Archival &Current Client CMS/DAM Bureau/Newsroom News Sports Fashion & Entertainment Sentiment Emotion Analytics Trailers Reviews Analytics Enriched Content in your CMS/DAM Advertising/Marketing Film/Television Social Media Public Domain Automated Source Collection & Alerting Content Creation Classification Methodology
  • 11.
    AI enhanced networkconnects knowledge to content creators
  • 12.
    ROI – Idea#1 AI Enriched Content – Upstream • Increased Content Generation = Incremental Revenue Increase • Automated Categorization & Extraction = Reduced Production Costs • Asset Harmonization = Discoverability • Disambiguated Knowledge Graph = Relevant Insights
  • 13.
    Idea #2 -Use an AI strategy to automate Ontology/Taxonomy creation. (classification strategy) Some definitions: Ontology : a conceptual model representing a specific field of information • Classes (e.g. “person”, “organization”, “dog”) • Attributes of classes (“age”, “birth date”, “foundation date”) • Relationships between classes (“marry”, “own”, “hire”) Taxonomy : is the science of classification according to a predetermined system – hierarchical • Taxonomy is the process of naming and classifying things such as animals and plants into groups within a larger system, according to their similarities and differences. Knowledge Base (KB) : contains the ontology model plus the instances of • Classes (“John Smith”, “Expert System”, “Buddy”), • Properties (“John Smith is 25 years old”) • Relationships (“Expert System hired John Smith”) • Adding instances to an ontology model = “populating” the ontology
  • 14.
    High-Level Workflow METADATA APPLICATIONS SEARCH ANALYTICS VISUALIZATION FACETS RECOMMENDATIONS PORTALS TOPIC PAGES LINKEDDATA KNOWLEDGE BASES ONTOLOGY DESIGN DATABASE/BIDOCUMENTS NLP ENGINE NLP DESIGN … Subject Matter ExpertSME End-User U CREATE ONTOLOGY MODEL 1. NLP ENGINE FEEDS YOUR ONTOLOGY 3. CREATE NLP PIPELINE 2.
  • 15.
    ROI – Idea#2 • Feeding your ontology/taxonomy is essential • AI strategy leverages semantic reasoning to : • Automatically feed your ontology/taxonomy based on your content • Resolve ambiguity for a comprehensive view of each entity • Maintain up-to date knowledge • The result ? A well-fed ontology/taxonomy • Provides end-users with relevant insights based upon knowledge extracted from your content
  • 16.
    Copyright © 2018Expert System - All Rights Reserved - Slide 17 Confidential – do not distribute Deep indexing and Data Extraction at • Goals : Dow Jones Professional Information Business • Drive a Compelling User Experience with Efficient Search & Alerting • Create a Differentiated Data Offering • The Solution : Expert System • Speed and Quality critical • Combining Classification and Extraction approaches • 11 languages • Classification • Applying Dow Jones Intelligent Identifiers • 3000+ Region/Topic/Industries • Extraction • First Area : Regulatory & Compliance : • Special Interest People & Adverse Media Entities Processing 1.5 Million Articles Every Day
  • 17.
    ‘‘ AI iscritical to our business’’ Bob Pashinsky Director, Content Strategy & Metadata
  • 18.
    Copyright © 2018Expert System - All Rights Reserved - Slide 19 Confidential – do not distribute Creating Meaningful Metadata at • Goals • Tag movie clips with meaningful metadata • Resolve cost and inconsistency of Manual indexing • The Solution : Expert System AI Capable Deployment • Harmonize terminology and clip tagging based on SONY term taxonomy • Real-time Taxonomy validation & enhancement • Indexers spend less time validating tags • Seamless integration with other platforms via Expert System REST API’s Overview
  • 20.
    AI : Stateof the Art  AI is here to stay: there are many problems that can be solved right now in a very effective way  To implement a solution based on AI, you need experts, time, analysis and hard work it is no different than anything else in the world  Some use cases which may seem to be a good fit for AI are not, cost analysis before implementation of any AI strategy may show that it is not cost effective
  • 21.
    AI after theHype  It is not the solution to all world’s problems and not the most important after the discovery of the wheel  It is not so advanced that a small amount of work can solve complex problems and it cannot learn without supervision;  Remember what Edison said about genius: “Genius is one percent inspiration and ninety-nine percent perspiration”
  • 22.
    AI after theHype  It is not machine learning or deep learning: ML is only a very common strategy used in many AI use cases  It is not an omniscient program that only needs to be configured or trained to solve real problems  It is not something that programs itself and functions without the help of experts which anyone can easily adapt to solve real problems
  • 23.
    AI : Stateof the Art  Implementing an AI solution always means programming a computer, in one way or another  In many cases, the best way to understand if an AI project can be a success is to do a Proof of Concept  AI is not the General Intelligence of human beings  An existing process doesn’t need to be necessarily replicated in the same way when AI is used
  • 24.
    Hints for aSuccessful AI Strategy  A good amount of healthy skepticism is always a good start to set the right expectations for AI  Have confidence in the value of AI and of experience: sometimes the road can be long but at the end the results are real, solid and measurable  Start with a gradual approach: targeting a step-by-step implementation of an AI Strategy is often the best recipe for success
  • 25.
    Thank You /Your Questions Ian Cameron, VP Media & Entertainment icameron@expertsystem.com 443 979 3869