IntelliCord

A Discord Bot to Track and Help Mental Health Challenges

Comment

Inspiration

According to Harvard Business Review, 76% of workers face some level of mental health issues at the workplace. This has been exacerbated as the pandemic has gotten worse. Lockdown has also moved people to increasingly online platforms like Discord. Thus the idea of IntelliCord came along. The main idea was to create an online debugger that helps programmers overcome the stress of programming and the debugging process. However, after a long thought, the team decided to take a more general approach and a more broad scope to support all workers. The idea of creating a Discord bot was to broaden our users, since we believed that making a webpage for this type of idea will not be as effective since people have to go through the process of opening the page and login in, etc. However through Discord, once someone added the bot they have easy access to the bot whenever they like since Discord is already a lot of people's main social platform.

What it does

IntelliCord is a Discord bot that analyzes chats from users and gives back a response based on the current mental state of that user, based on metrics from 8 emotions. For example, if someone is sad they might want to hear a joke or two and that's what the bot will do. The bot will also give out suggestions/support depending on the current mental state. If the user is sad, the bot might suggest how to overcome it and even notify everyone on the text channel to help with their problems. If the user is happy then the bot might suggest a quick break or a word of support for the user. Other than analysis, we also implemented the bot to be able to text back to the user if the user needed a bot to talk to.

How we built it

We used co:here as a tool to train an NLP Large Language Model (LLM) to determine the current mental state of a person and use machine learning to create a priority based on how long a chat has been sent. We combined five large datasets for sentiment analysis to train through co:here. Through both of those, the bot will then determine the output for the user based on the result of the analysis. For the output, we used GPT-3 to generate the response that is suitable for a response of a certain situation. We switched our LLM to GPT-3 because we found that it consistently provided more suitable responses than co:here, though we did try co:here text generation.

What's next for IntelliCord

We think that our next steps should include improving the AI itself to be broader with more emotions than the current 8 emotions that it has now as well as increasing its accuracy since adding more labels might decrease the AI's accuracy. We would also like to integrate certain mental health tools into the bot such as a Pomodoro timer, or a calming in-chat game. We also think that IntelliCord should be expanded to other social platforms to reach a broader community and help many people around the globe.

Built With

Share this project:

Updates