What it does
m.path is a fast and reactive Discord chatbot specializes in identifying emotions expressed in conversations between the user and a correspondent. The user may pass messages of any size as input to the chatbot and, using GPT-2 and Multinomial Naive Bayes, returns a list of likely emotions the sender was expressing within them.
How we built it
- GPT 2 for processing input
- Multinomial NB model trained on dataset for identifying emotion
- Hosted on Discord (attempted Wordpress and Github Pages) ## Challenges we ran into
- GPT-2 self-looping text generation (solved by using a Multinomial NB for increased accuracy in emotion detection)
- Multinomial NB model somewhat biased towards detecting joy
- WordPress website hosting difficulties
- Github Pages Python integration difficulties ## Accomplishments that we're proud of
- Mastery over GPT-2 API
- Effective use of LLMs for Natural Language Processing
- High accuracy in written emotion detection ## What we learned
- Using GPT-2 API for text processing, generation, and context identification
- Using Multinomial NB for emotion identification
- Training + verifying accuracy of models
- Using Discord API for reading and processing user input as well as writing output
What's next for m.path
- Implementation across multiple social media platforms
- Increased scalability of model(s)
- Understanding of complex linguistics (e.g. sarcasm, slang)
Log in or sign up for Devpost to join the conversation.