Enterprise Chatbots: A Hybrid Approach to Machine Learning

18
www.creativevirtual.com Enterprise Chatbots: Taking a Hybrid Approach to Machine Learning Björn Gülsdorff, Head of Business Development @Bjoern_CV @creativevirtual

Transcript of Enterprise Chatbots: A Hybrid Approach to Machine Learning

Page 1: Enterprise Chatbots: A Hybrid Approach to Machine Learning

www.creativevirtual.com

Enterprise Chatbots: Taking a Hybrid Approach to Machine Learning

Björn Gülsdorff, Head of Business Development

@Bjoern_CV @creativevirtual

Page 2: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Gartner Predicts

Source: Gartner, 2016

Page 3: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Existing VAs – how to add ML?

Page 4: Enterprise Chatbots: A Hybrid Approach to Machine Learning

The challengeof

Enterprise Chatbots

Page 5: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Recap of the Challenges

Integration

Customisation – Domain Knowledge

Compliance (Legal, Security, …)

And ROI Timeline Budget

And Predictability

Page 6: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Self-Learning - Pitfalls

Machine Learning is Not flexible• Architecture

• Algorithm

• Special cases

Black Box –By design

Garbage in –Garbage out

Page 7: Enterprise Chatbots: A Hybrid Approach to Machine Learning

And: A noisy Panda is a Gibbon

Panda (57%) „Noise“ Gibbon (99%)

Page 8: Enterprise Chatbots: A Hybrid Approach to Machine Learning

What‘s missing?

Let‘s think!

Page 9: Enterprise Chatbots: A Hybrid Approach to Machine Learning

If your agents were quicker

readers…

Would you let them

read chats or would you train them?

Page 10: Enterprise Chatbots: A Hybrid Approach to Machine Learning

If your agents were multi-taskers…

Should they watch other agents or would you train them?

Page 11: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Accept Human Knowledge!

Page 12: Enterprise Chatbots: A Hybrid Approach to Machine Learning

How it worksUnsupervised pre-learning of generic language model

System actions with integration (Classes) defined by humans

Machine Learning against examples

Scenarios defined in flow editor

Examples for classification defined by humans or collected from agent actions

Page 13: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Hybrid2 Approach

1. Creation of Content• Unsupervised Pre-Learning

• Human curated content

• Supervised Learning

2. Improvement• Alternates from user

interactions

• Feedback from agents

• Retrain

Page 14: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Finds & Benefits

Small receptive fields

Robust against parameter variations

Only few examples needed

Great results!

Page 15: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Semantic

StatisticalSurveys

FeedbackLoops

HumanCuration

A hybrid approach of human curation of content and self-learning minimizes the risk in creation, allows the system to continually improve while also giving control over the reliability of virtual agent responses.

Page 16: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Companies needtoday

Combination of virtual & real

Knowledge management & feedback loop

Complementary technology

Learning capabilities

Page 17: Enterprise Chatbots: A Hybrid Approach to Machine Learning
Page 18: Enterprise Chatbots: A Hybrid Approach to Machine Learning

Get in Touch with Me

By email:

[email protected]

On Twitter:

@creativevirtual

On the web:

www.creativevirtual.com