TL;DR
Chatbots are a very helpful and innovative way to help users solve common issues, but building it takes a lot of time & planning.
Session Notes
Yes, Dave, I CAN Do That: Building a Canvas Support Chatbot
ErinGreen |Senior Project Management Specialist | Northwestern University
pgo586 | Senior Software Developer | Northwestern University
Jonathan Diehl | Senior Blended Learning Specialist | Northwestern University
Intro to Chatbot
Why build a chatbot?
What if 24/7 Canvas support went away? Also a way to work with IBM Watson.
User asks goes to chatbot then back to info.
Asks questions to narrow the path in the decision tree.
Implementation
Intent- the users intention when coming to the chatbot.
High level analysis of tickets for creating ground work of the chatbot.
Classified the common issues to determine users intents.
Top10 Intents of about 170.
First version focused on the top 50 intents.
Created the sample. Made 10-20 for how a user would state the intents.
4 main responses
- Direct response
- Decision tree
- Recommended reading
- Escalation
Architecture
Use google search when it doesn’t know answer.
Added to ticketing system.
User interface.
Chatbot LTI located in the Help menu.
The inferential knowledge is optional but helpful. This is determining role from context of the question.
Having dynamic learning will help the bot learn for the user responses.
Devices used
- Botkit
- AWS services
- Dashboard lite - used for analytics.
Lessons Learned
Where is it now? In the Canvas help menu! Soft launch for fall.
*All of the content here is from my live session notes or straight from my memory. This means that I may miss something or be mistaken in my content. Please feel free to add to, put in context, or correct any mistakes by using the comments below.
