1 of 14

From People2Vec to Emotional AI

www.people2vec.org

We don’t see things as they are. We see things as we are ! - Anais Nin

Jan 2016

2 of 14

Understanding H2H Communication

3 of 14

Content Centric

Siri, Cortana, Alexa, ...

  • Content Centric
    • Question - Answering
    • Light dialog
  • Context Sequence(s)
  • Knowledge or Actions

Far from the Human communication

HER, Sarah, HAL (or not ;p)

  • People (person) Centric
    • Human like dialog
    • Empathy & Emotion
  • Global Context
  • Concept Learning

Human emotional communication

People Centric

4 of 14

Why Emotions ?

Love, love, love …

… All you need is Love !

Photo de Buddy

Source : https://www.pinterest.com/pin/415527503094644660/

http://www.bluefrogrobotics.com/en/buddy/

Create Emotions to enable Empathy and facilitate adoption of Virtual Assistant !

5 of 14

How: Language, Emotion & Personality !

6 of 14

How to infer Emotion ?

The role of Language in the Construction of Emotions

Model : Vector Space

belgium

wallonia

luxembourg

netherlands

Word2Vec : Word Embedding with Context

barack_obama

joe_biden

hillary

happy

glad

afraid

Language => Sentences => Phrases => Words �Context : Words are characterised by the company they keep

7 of 14

What : People2Vec ...

Capture and adapt to People Language

  • People BOW (PBow)
    • Twitter, Facebook, Linkedin, ...
    • Docs, Emails, Blog Posts, …
  • People Image Captioning
  • People All Things Data (IoT & Services)

People as Context Vectors

  • Doc2Vec(PersonalBOW or PeopleBOW)
  • Similarity, CBOW, Skip-Gram, Topics, S2S, ...

Deep Learning : Sequences & Sentiment

  • Absolute and Relative Analysis

Things Data

belgium

wallonia

luxembourg

Herseaux

karate

sport

Tennis

Deep-Learning

machine_learning

word2vec

Personality - Values - Needs - Moods - POIs … �with Deep Learning

8 of 14

Applications of People2Vec ?

People as a Vector Space (Bag-of-Contexts) to be used in Deep Learning

Person Similarities

  • Person 2 Person
  • People 2 People

People Classification

...

People 2 Task

People 2 Knowledge

People 2 Entities (enterprise, R&D Papers, …)

...

Sentiment Analysis

Time-Series Classification

...

9 of 14

Read and post for me on twitter !

(Please) I’m tired !

-

People2Vec + Twitter

10 of 14

Keep an eye out for me !

P2V(Me) reading assistant :

  • Input :
    • Twitter Text History
    • LinkedIn Text Profile
    • Feed ~ 300k news / day

11 of 14

Keep an eye out for me !

P2V(Me) reading assistant :

  • Input :
    • Twitter Text History
    • LinkedIn Text Profile
    • ~ 300k news / day�
  • Output :
    • < 10 top news qualified / day
    • > 98% skills accuracy

12 of 14

Use Cases ?

Dialog & Discussion

Teaching & Learning

Mood Swing Management

Enrollment

Enterprise Communication

Staff Management

Re-Identification

Self Management

Dating

Semantic Search

Health Care

Personal Coaching

Depression

e-Commerce

Reading

Car Personal Assistant

Social Robots

Things2Vec

Call-Center

Social Personal Assistants

13 of 14

People2Vec.org : Join us to build it !!!

  • The Virtual Assistant is the new website !
  • Deep Learning & Word Embedding can outperform the current tools
  • Join our community : www.people2vec.org
    • Audrey Duet, Louis Monier, Greg Senay, Fabien Lavocat & Greg Renard
  • Find out who we are :
  • Please, when creating Virtual Assistants and AI solutions, do it with Ethic. Always!

14 of 14

References