Sites like Twitter have given people a platform to share their opinions and give the people without a voice the ability to speak. However, the overall quality of social media such as Twitter is reduced by the amount of spam and hate speech.
What it does
Our application, Kite AI seeks to eliminate the rising levels of abuse and harassment in Twitter. While Twitter and other social media companies are rolling out some of their own devices in attempts to provide a safer, healthier environment, we upped the stakes by using machine learning, combined with the ability for users to go in and double check our algorithm's decisions, to let it ultimately become a personalized feed curator.
How we built it
We utilized Bayesian learning and Python's Natural Language Toolkit library for the actual algorithm, isolating 5 patterns in language which serve to provide a greater probability of "low quality" or abusive tweets. By turning our initial algorithm into an API hosted on AWS, we then created a dashboard which displays tweets that were blocked by our code, but flagged as a questionable (or "low confidence") block.
Challenges we ran into
Originally, we wanted to write this as a neural network type of deep learning. The fact is, many people have tried that, and all have failed for the same reason. There isn't a deep enough dataset that lets a neural network respond accurately enough to be acceptable. After discussion, we decided to switch to a more statistical approach by using Bayesian learning- which we ultimately kept.
Accomplishments that we're proud of
We are very proud of the accuracy of our algorithm, given the limited data that we could use. We had to create our own dataset for abusive and non-abusive tweets, which when used in conjunction with our 5 variables provides a very accurate response without any user data. We are also immensely proud in our website design.
What we learned
Each member of our team came to hackGSU with an idea to work with something we had never touched before. Machine learning, data sciences, and web animations were just the start of what we learned during our hacking.
What's next for Kite AI
The next step for Kite AI would be to sit and gather crowd data through users. The more users that use our dashboard, the more accurate our algorithm will become, eventually having the potential to correctly analyze sarcasm and friendly humor.