When privacy matters! A series of challenges for chatbots in data-sensitive businesses such as healthcare and finance by Christophe Willemsen
Meetup: Integration of Chatbots in Healthcare and BFSI, Dubai, 1.11.2018
12. ● Creating a Chatbot with Deep Learning, Python, and TensorFlow
● How Chatbots Are Learning Emotions Using Deep Learning
● How I Used Deep Learning To Train A Chatbot To Talk Like Me
● Neuralconvo - Chatting with a Deep learning brain
● ...
Current conversation topics
13. ● Creating a Chatbot with Deep Learning, Python, and TensorFlow
● How Chatbots Are Learning Emotions Using Deep Learning
● How I Used Deep Learning To Train A Chatbot To Talk Like Me
● Neuralconvo - Chatting with a Deep learning brain
● …
● Training your Machine Learning models on the blockchain
● Conversational Artificial Intelligence in the FinTech
Current conversation topics
19. ● It is the identification and categorization of what a user online
intended or wanted when they typed their
● Intent detection using semantically enriched word embeddings
● Zero-shot User Intent Detection via Capsule Neural Networks
● A Bi-model based RNN Semantic Frame Parsing Model for Intent
Detection and Slot Filling
Intent detection
34. Coreference Resolution
● Bot : Hi, I am your health assistant, how can I help you ?
● User : I have pain in the neck, it is really burning
35. Coreference Resolution
● Bot : Hi, I am your health assistant, how can I help you ?
● User : I have pain in the neck, it is really burning
Refers to
39. Sentiment Analysis
● Sentiment analysis aims to determine the attitude of a speaker,
writer, or other subject with respect to some topic or the overall
contextual polarity or emotional reaction to a document,
interaction, or event
42. Sentiment Analysis
● If chatbots are not your primary mean of communication, it
means you failed at other points ( the user could not find the
information online )
● A typical negative, neutral, positive polarity does not reflect the
reality
45. Expectations vs reality
● Bot : I am your financial assistant, how can I help you ?
● User : Hi, I could not login into the online banking system
46. Expectations vs reality
● Bot : I am your financial assistant, how can I help you ?
● User : Hi, I could not login into the online banking system
● The app doesn’t work, dunno man, wtf ?
47. Expectations vs reality
● Bot : I am your health assistant, how can I help you ?
● User : Hi, I have a cough
48. Expectations vs reality
● Bot : I am your health assistant, how can I help you ?
● User : Hi, I have a cough
● ﺳﻌﺎل ﻋﻧدي ، ًﺎﻣرﺣﺑ
49. Expectations vs reality
● Bot : I am your health assistant, how can I help you ?
● User : Hi, I have a cough
● ﺳﻌﺎل ﻋﻧدي ، ًﺎﻣرﺣﺑ
● Most NLP Processing techniques for English do not work for
Arabic
● Arabic has 40 dialects
● POS tagging is dependant on the domain
● etc...
51. Data-privacy is a trade-off
● When you make the choice of data-privacy, you have trade-offs
○ You do not sit on the shoulders on giants anymore
○ The cost of research will exponentially surpass the usage cost
of a third-party solution
○ You have the sole responsibility of the satisfaction of the
customers
55. Integrate all the things
● Intent Detection
● Natural Language Processing
● Custom Named Entity model training
● Co-reference resolution training
● English, Arabic, German support
● Graph-based context builder