oyoty - lessons learnt building a “chatbot” for children
TRANSCRIPT
Lausanne Chatbots meetup François Helg - 1st March 2017
Lessons learnt building a “chatbot” for children
WHAT ARE WE TALKING ABOUT
WHAT ARE WE TALKING ABOUT
Jun 2015
Mar 2017
Sept 201
5
MVPFeed-
back 1
Oct 2015
Conversational UI + Engine MVP
Jun 2016
Feed- back 2
Jul 2016
Lessons learned
Improved conversations Chatbot experiments
Facebook Messenger supports chatbots
Apr 2016
Challenges + Decisions made
The world’s first online digital assistant for children
FEEDBACK ON FIRST MVP
INSPIRATION
➤ Not so much by the conversational UI and ChatBot trend as a whole (was a bit before the hype)
➤ By a particular application
➤ Quartz
CHILDREN LOVE CHAT APPLICATIONS
Source OFCOM: Children and parents: Media use and attitudes report 2015
TRADITIONAL TEAM ORGANISATION
UX DesignerEngineers
Team
Product Owner
LEVEL OF SOPHISTICATION
➤ Transactional and narrow domain
➤ Incident resolution
➤ Based on a decision tree
➤ The child follow a pre-defined scenario
https://medium.com/convcomp2016/conversational-vs-transactional-chatbots-172166f3d2fc#.nfl0bbbb5
CREATE A PERSONA FOR OUR BOT
A
1 2 3
B
C
https://chatbotsmagazine.com/your-chatbots-personality-is-the-key-683a2051fb7d
SCENARIO DEFINITION
ADD ONE MORE PROFILE
UX Designer
Engineers
Team
Product Owner
Content writer
CONVERSATION - CREATION / REVIEW PROCESS
FEEDBACK ON OUR CONVERSATIONAL UI
➤ Let me come back if I’m not sure!
➤ Be more funny and less predictable
➤ What if I could ask questions to the bot
LET ME COME BACK IF I AM NOT SURE
➤ Conceive the conversation as threads or topics
➤ Better abstraction of human conversations
➤ People want to explore what is possible
BE MORE FUNNY AND LESS PREDICTABLE
➤ Add emotion to the bot
➤ Support for more multimedia content : image, animated GIF, emoticons, videos
➤ Add some unpredictability in the formulations
➤ Randomness, keep more context into consideration
WHAT IF I COULD ASK QUESTIONS TO THE BOT
➤ Trigger conversations based on known categories
➤ Simple NLP to understand intent in questions
➤ “Tell me about cyberbullying”
➤ Evaluating API.ai (if you have a feed-back to give us, please go ahead)
INVEST IN MACHINE LEARNING
➤ We are working with an EPFL lab to work to help our bot match answers to elaborate questions
➤ “What should I do!!! I just shared an image of myself naked to that asshole of Chris, my ex boyfriend.”
EXPLORING CHATBOTS
➤ On platforms where our users are
➤ Kik, Facebook
➤ For particular use cases
➤ Report Cyber-bullying
➤ Provide educational content
➤ A channel to acquire new users
SUMMARY
➤ Create a personality for your bot
➤ Start with a narrow domain, solve a real problem and learn
➤ The quality of the conversations is key
➤ Find a good content writer
➤ Add image, video
➤ Add some unpredictability
➤ Let people explore what your bot can do
➤ Keep an eye on the ecosystem
INTERESTING RESOURCES
➤ Reading
➤ ChatbotMagazine
➤ Virtual conference
➤ http://www.botscamp.co/
➤ Meetups
➤ Lausanne Chatbots
➤ Swiss Chatbot