o-gent / facebook-hackathon

Messenger competition mental health bot

drones drones drones drones drones

facebook-hackathon's Issues

test cases

Need some more cases of the bot not working correctly. Try anything you can think of messaging the bot and see how it reacts. If you think it reacts wrong, post a screenshot below!

We can then use this to fix errors and improve the AI detection

Send multiple messages in succession

Can we send multiple messages in succession?

for the response/fb_message() function can we pass a list of messages, this function then sends the messages one after the other?

https://stackoverflow.com/questions/47483190/sending-multiple-reply-messages-on-single-postback-in-facebook-messenger-bots

• create a new branch for the feature

• let fb_message() accept a list of messages and do the correct type hints

• for a single message a list with a singe string should be passed instead

• modify the data payload to include multiple messages? or send multiple post requests?

• create a merge request to master for review

default_intent

Shouldn't be possible to get default intent as an answer. Perhaps ask the user to describe in a different way again if we get this reply from wit?

Happy / Sad was incorrectly identified

fix going from end state to start state again

Bot doesn't go back to start state once it ends :(

We need a name

We need to decide a name for our creation

add your suggestions below!

Person class fix I/O

I/O should be programmatic

• All methods need to return the text output wanted, rather than printing

• All methods should accept a string variable as the input (no input())

• Testing function should then handle input() and printing results

User conversation rating

User rates the experience at the end of the conversation, so we can improve user experience

• new state at the end of the conversation to send quick reply ratings

• assign this rating to the sequence of events they went through.. somehow...

stress branch implementation & testing

Created to track progress of testing on these features:

• correct diagnosis of stress in conversation

• correct diagnosis of stress reason in conversation

• suggestions to combat stress reason

Follow-up message

Use the one-time reply feature to send a follow up message a month later to see how the user got on with what we suggested

• research API requirements

• implement api interface

• implement control flow into app

• test

more natural conversations

At the moment I feel it's a bit disjoined asking how they feel, then asking if they are stress (as well as the possibility we get it wrong)

• when we ask the user how they feel, try and detect whether they are e.g. stressed as well as what the stress reason is at the same time.
• We then ask if they are stressed, if we have detected this. Otherwise ask again.
• We then ask if they are e.g. tired, if we have detected this. Otherwise ask them to explain the reason why they are stressed.
• Maybe ask them to confirm if they are having problems with sleep.

Would this work? would it be better?
It uses what we have already, just rephrasing into hopefully a more natural experience.