Slide 1

Slide 1 text

Conversational AI: Building clever chatbots Tom Bocklisch, Lead ML Engineer @ LASTMILE

Slide 2

Slide 2 text

Example of a live Skill: A customer can change her address via Facebook Messenger Conversational AI will dramatically change how your users interact with you.

Slide 3

Slide 3 text

An open source, highly scalable ML framework to build conversational software

Slide 4

Slide 4 text

is a technology company developing conversational AI. Goal: next-generation intelligent bots Team: tight-knit, fast-moving team of researchers, engineers, designers and product people Location: everywhere (honestly: Berlin, Edinburgh, Beijing) We work on the core technology for next-generation conversational AI Founders: Dr. Alan Nichol (CTO) Alexander Weidauer (CEO) Advisory Board: Chad Fowler (MD & CTO @ Wunderlist) Matthaus Krzykowski (former Co-Founder @ Xyo) Cat Noone (Designer & Founder @ Iris) Investors: Reference customers: Introduction

Slide 5

Slide 5 text

Architectural Overview Introduction AI (Natural Dialogue Management) (Natural Language Understanding) (Natural Language Generation) (Conversational Platform, e.g. Facebook Messenger) “What’s the weather like tomorrow?” (User Request via text or voice) “It will be sunny and 20°C.” (AI response via text or voice) (Your backend, database or API)

Slide 6

Slide 6 text

Under The Hood

Slide 7

Slide 7 text

Natural Language Understanding Natural Language Understanding What’s the weather like tomorrow? Example Entity Extraction Pipeline ”What’s the weather like tomorrow?” { “date”: “tomorrow” } Tokenizer Part of Speech Tagger Chunker Named Entity Recognition Entity Extraction Example Intent Classification Pipeline ”What’s the weather like tomorrow?” { “intent”: “request_weather” } Vectorization Intent Classification Under The Hood

Slide 8

Slide 8 text

Demo

Slide 9

Slide 9 text

1. Create your training data Demo E.g. using the contributed rasa NLU gui at https://golastmile.github.io/rasa-nlu-trainer/

Slide 10

Slide 10 text

2. Configure the model Configure the model Demo

Slide 11

Slide 11 text

3. Train Training the model Demo

Slide 12

Slide 12 text

4. Use Model Playing around with the trained model Demo

Slide 13

Slide 13 text

Under The Hood

Slide 14

Slide 14 text

Dialogue Handling Under The Hood “What’s the weather like tomorrow?” Intent Entities next Action State previous Action “Thanks.” after next Action updated State “It will be sunny and 20°C.” SVM Recurrent NN ...

Slide 15

Slide 15 text

Detailed Dialogue Handling Under The Hood “What’s the weather like tomorrow?” “It will be sunny and 20°C.” Entity Input Action Mask Renormal- ization Sample action Action type? Response API Call Recurrent NN API Call Entity Output Intent Classification Entity Extraction Similar to LSTM-dialogue prediction paper: https://arxiv.org/abs/1606.01269

Slide 16

Slide 16 text

Final Thoughts

Slide 17

Slide 17 text

Open Challenges Final Thoughts Challenges for curious minds: ● ● Unsupervised multi-language entity recognition ● Dialogue generalisation (e.g. optional questions) Chit-Chat model Task-Oriented model I want to travel to Spain. ? ● Combination of different dialogue models

Slide 18

Slide 18 text

Current Research Final Thoughts Good reads for a rainy day: ● Last Words: Computational Linguistics and Deep Learning (blog) https://goo.gl/lGSRuj ● Memory Networks (paper) https://arxiv.org/pdf/1410.3916 ● End-to-End dialogue system using RNN (paper) https://arxiv.org/pdf/1604.04562.pdf ● MemN2N in python (github) https://github.com/vinhkhuc/MemN2N-babi-python

Slide 19

Slide 19 text

● Conversational AI is a big part of the future ● Deep ML techniques help advance state of the art NLU and conversational AI ● Open source is strategically important for enterprises implementing AI Summary Final Thoughts 3 take home thoughts:

Slide 20

Slide 20 text

Get in touch! Tom Bocklisch Lead ML Engineer [email protected]