Inspiration
Ru moved from the GTA to the San Francisco Bay Area four years ago for school. During that time, he kept in contact with his brother Roy specifically via an online FPS called Valorant. The relatively toxic community prompted the creation of Toxic, which is applicable to games, social media, online forums, and more!
What it does
Toxic is a proof-of-concept that aims to detect and notify users about toxicity. It uses Coherent's NLP API to use sentiment analysis to calculate a toxicity score (between 0 and 1, with 1 being the most toxic), and Twilio's SMS API to notify users upon request. It currently has the following features:
Twitter Handle Search
Users can search any twitter handle and get an analysis of the Twitter user, their followers, and their following's toxicity
Twilio Notification
One can stay up-to-date with anybody they searched for by opting-in to text notifications whenever that handle, their followers, or their following exceeds a toxicity threshold of 0.8
Aribtrary Text Evaluation
Toxicity Analysis of an arbitrary conversation via text file upload
Statistical Analysis
B2B oriented statistical analysis of toxicity score (see below)
How we built it
The project is entirely coded in Python. The GUI is created with the help of Tkinter. Additional backend Jupyter Notebook code is available that includes statistical analysis, which includes assessing an arbitrary confidence interval for the true toxicity score by modelling it with Student's T-Distribution and creating a Bayesian posterior probability density function given the data and a Uniform prior.
What's next for Toxic
Immediately, we hope to flesh out the GUI and link all functionality. Over a longer time horizon, there's much opportunity for more statistical analysis and visualizations. We also want to natively integrate from more platforms (i.e. social media). Finally, we hope to add a voice feature where there's a voice-to-text bridge that facilitates the evaluation of audio files.
Log in or sign up for Devpost to join the conversation.