Hack the Hack

Inspiration

What it does

The project aims at modelling and analyzing all the submissions on devpost to provide a better understanding of what it takes to win a hackathon. We take the semantic information from the description of the project, in combination with team information like the number of team members, feed them into a neural network, and predict the possibility of each project securing an award. Some interesting findings include that apps and games are the most commonly tackled subjects, big teams are more likely to win, and javascript is the most popular language in hackathons.

How we built it

Scraped the whole DevPost website as our dataset; tokenizing, stemming, stop word filtering every project's tagline; turning it into a 100 dimensional vector; in combination with team size, fed into a binary classification neural net after balancing/scaling/centering; and finally produce the prediction of winning probability for each project.

Challenges we ran into

Accomplishments that we're proud of

The project created a paradox by predicting itself having only 5% winning probability. By either winning the hackathon or precisely predicting its own failure, it proves itself.

And guess what, it didn't get into the finalist of this hackathon.

So we guess in a way, it won. xD