Dis Care

A Smart Discord Bot That Has Your Back

Comment

Inspiration

Mental Health is often overly stigmatized, making people hesitant about seeking out help, but I've seen people be more open about ranting or talking about their problems online, due to the ability to be anonymous online. Such social media platforms like Discord offer people the chance to talk about these topics but often their small, side remarks go unnoticed. The Bot DisCare would offer them the chance to have their problems addressed.

What it does

It screens through discord messages and, using a BERT model I'd trained on a dataset of tweets, determines how likely a messages is to be someone expressing emotions of anxiety, overly stressed, or otherwise needs help. It'll then message them words of encouragement.

What we did

BERT and Hugging Face were used to tokenize and train a dataset of roughly 3000 “normal” and 3000 “disturbed” tweets.

Discord.Py library was used to control the discord bot.

Challenges

The most time consuming part of the entire challenge was figuring out how to get the model working. This came with so many failures since this was my first time doing AI. For instance, almost all of day 1 went to waste after I struggled to open and read a massive dataset (total of about 8 GiBytes once converted into CSV) only to realize it didn't contain what I wanted. Not to mention all the little bugs that came up and cost me hours each time. Overall I had to overcome code regularly failing, not knowing how to handle datasets, and needing to restart twice all while trying to learn NPL.

What we learned

This is my first time working with NPL and making a discord bot so I learned a lot about doing it using google collab. This was my first big coding project so I’m proud that I managed to get it to work and got through all the difficulties.

What's next

Due to an unbelievable amount of setback and coding challenges, it ended up not being fully finished on time so I hope to add the feature of sending a list of helpful resources, hotlines, and motivational quotes. I also would like it to have the ability to check-in on users who may have said things repeated.

Built With

Share this project:

Updates