From Chatbots to Augmented Conversational Assistants: An Experimental Study combining AI and Crowdsourcing
Developing technology, including Natural Language Understanding and Machine Learning are taking Chatbots, Virtual Assistants, Conversational Bots… to an extent unheard of in the past. And yet, we cannot but notice that efforts are still needed for these interactive agents to attain a satisfactory level of conversational abilities. As a matter of fact, Bots still often cause a classic frustration effect when they reach limits in the scope of their powers.
In order to take Bots beyond their technical boundaries:
We augmented them with technologies such as advanced data processing and Artificial Intelligence, including computer vision specific capabilities to mention just a few.
We designed a unique approach based on crowdsourcing methods complementing and enhancing existing techniques to AI / Deep Learning.
In this talk, we will present the result of our research based on two tracks:
As part as our first iteration of research (aka “AVA”), we focused our efforts on identifying the most relevant AI/ML features that bring advanced conversational capabilities to Chatbots.
In a second iteration (aka “CIVA”), we concentrated our research on bringing more understanding and replying abilities to Bots defining a progressive methodology where data are improved.
In this presentation you will see how we augmented Chatbots capabilities with:
Machine Learning
Computer vision
Sentiment analysis
Crowdsourcing
Conversational methods
From Chatbots to Augmented Conversational Assistants
1. From Chatbots to
Augmented
Conversational Assistants:
An Experimental Study combining AI and Crowdsourcing
Frederic Jacquet
Chief Innovation Officer
Lao-dja S. Tchala
Data-Scientist
2. Agenda
§ Talk Overview: What can you expect
§ AI/ML to Augment Chatbots
capabilities
§ Methods & Human to improve bots
knowledge base
§ Perspective: Making an Inclusive
Chatbot
§ Wraping Up
3. • Talk Overview: What can you expect
• AI/ML to Augment Chatbots capabilities
• Methods & Human to improve bot knowledge base
• Perspectives: Making an Inclusive Chatbot
• Wrapping Up
4. "By 2020, the average
person will have more
conversations with bots
than with their spouse"
Gartner 2016
Gartner's Top 10 Strategic Predictions for 2017 and Beyond: Surviving the
Storm Winds of Digital Disruption
5. An Experimental Study combining AI
and Crowdsourcing
Identifying the
most relevant AI/ML
features that bring
advanced conversational
capabilities to Chatbots.
Bringing more
understanding and replying
abilities to Bots defining a
progressive methodology
where data are improved
based on crowdsourcing.
• Methods and Human
• AI/ML to Augment Bots
7. • Talk Overview: What can you expect
• AI/ML to Augment Chatbots capabilities
• Methods & Human to improve bot knowledge base
• Perspectives: Making an Inclusive Chatbot
• Wrapping Up
8. AI : Computer Vision –
multiple choices
• What is it?
• Why?
• Our hypothesis
• Which technologies?
9. AI : Computer Vision –
multiple choices
• How it works?
10. AI : Computer Vision – multiple choices
Train a model with a set of images
1 Track a hand on the images
2 Calculate the center of the cercle around the hand
3 Coding an action depending on the center’s position
4
ACTION 1
ACTION 2
16. ML : Recommendation algorithm
• General
Qi
Qj
Qk
Full Text
Full Text
Full Text
Ai
Aj
Ak
Ai
Aj
Ak
Ai Qi
Aj Qj
Ak Qk
Qi
Qk
Key_word_i1 Key_word_i2
Key_word_k1 Key_word_k2
User send message
1 Bot answers
2 Mapping
3 Key words extraction
4 Model prediction
5
Qi
Qk
Ai
Aj
• Helpdesk use case : equipment's recommendation
Qj
Full Text
Full Text
Full Text
Qi
Qk
Manager
Equipment_i1 Equipment_i2
Equipment_k1 Equipment_k2
Authentificated users send messages
1 Key words extraction
4
Developer
Administrator
Qi
Qk
2 3 Rôle prediction
5
Administrator
Developer
Administrator
tips
Bot predicts the user’s rôle and proposes supplement equipments
6
Equipment_i1
Equipment_i2
You must be an Administrator
You might also need
Administrator
20. • Talk Overview: What can you expect
• AI/ML to Augment Chatbots capabilities
• Methods & Human to improve bot knowledge base
• Perspectives: Making an Inclusive Chatbot
• Wrapping Up
30. Adaptive answers
Qi
Qj
Qk
Full Text
Full Text
Full Text
Ai
Aj
Ak
Ai
Aj
Ak
Ai
Aj
Ak
Ai
Ak
User send message
1 Bot answers
2 Training Data from user
3 Model predict which answers the user may like or not
4 Model training
5
Aj
Al
Ak
New answers’ prediction
True values Predcitons
Memory to use in next conversations
Al
Ak
6
36. Reformulating answers
Qj
User send message
1
Ai
Ai1
Ai2
Ai3
Ai Ai1 Ai2
Ai3
Bot chose a response and send it to the user
2 The already given responses are removed from the ones the bot will chose from
3
Qj
Ai
Ai1
Ai2
Ai3 OR
4
42. Ask for clarification
Qi
2 questions are similar
0
Qi
Qi1
Qi1
Qj
Qj1
Qj1
Qi
Qj
Qf
Qe
Qf
User ask one of them
1
Key_word_i1 Key_word_i2
Key_word_k1 Key_word_k2
Key_word_i1 Key_word_i2
Key_word_k1 Key_word_k2
Keys words are extracted from each of those questions
2 Keys words are used to ask clarification to the user
3
Do you mean
Key_word_i1 Key_word_i2
Key_word_k1 Key_word_k2
or
Key_word_i1 Key_word_i2
Key_word_k1 Key_word_k2
?
46. • Talk Overview: What can you expect
• AI/ML to Augment Chatbots capabilities
• Methods & Human to improve bot knowledge base
• Perspectives: Making an Inclusive Chatbot
• Wrapping Up
47. Inclusive Chatbots
Because texting or
talking may prove to
be temporarily or
permanently
challenging because of
morphology specifics,
trauma, mobility
deficiency, stroke,
diseases or any
deficiency.
48. • Talk Overview: What can you expect
• AI/ML to Augment Chatbots capabilities
• Methods & Human to improve bot knowledge base
• Perspectives: Making an Inclusive Chatbot
• Wrapping Up
49. Feedback / Questions
Do you have any questions?
Your feedback is important to us.
Don’t forget to rate and review the sessions.