Mental health advocacy is something our entire team is very passionate about. We wanted to use this opportunity to prototype a solution that addresses the need for quick, short-term self-help services based on the principles of cognitive behavioural therapy. And thus the Moodverse mental wellness chatbot was born!
What it does
Say you're experiencing a low mood and want to talk it out with Moodverse. Moodverse will use natural language processing and your conversation to identify the severity of your low mood and the possible sources of stress or anxiety leading up to it.
Based on that, it will try and get you to identify cognitive distortions, that is, patterns of negative thinking, and talk you through reframing them to help you feel better!
Moodverse also gives you insights over time about your mood and the common negative patterns of thinking you've had in the recent past.
How we built it
We used a Firebase and Python for the back-end with several machine learning models for classification and generation of text.
Sentiment Analysis Model
- Used Google Cloud's NLP API to gain insights about how users feel about certain stressors they talk about.
- Generated a self-love score to know how users were feeling about themselves during a session
- Used generative models to come up with sympathetic and encouraging statements.
Cognitive Distortions Model
- Used cosine similarity between sentences with cognitive distortions and new input sentences to detect and classify common cognitive distortions.
- Used Bootstrap Studio for a web-based application
- Chat-box interface for communication
- An insights section for analytics
Challenges we ran into
- Data mining and model training took a lot of time and debugging, but it was well worth it!
- Building a user interface as a team of back-end developers.
Accomplishments that we're proud of
- Scraping, processing, and training on tonnes of data to gain cool insights!
- Lots of solid natural language processing models, pre-trained and self-trained
- A lovely UI
- Using cool tools such as Google Cloud NLP API, Boostrap Studio, and Keras.
What we learned
- Machine learning = data, data, data!
- Integrate the back-end and front-end early on
What's next for Moodverse
- Gaining more input from healthcare specialists
- Integrating voice-to-text features