What it does
An essay grader that is built to rate essays and provide a variety of stats on your essay.
How we built it
We heavily modified an old essay rating artificial intelligence model from the The Hewlett Foundation: Automated Essay Scoring competition to suit our needs. This included changing the scoring system so it became a percentage grade (wrecking havoc on the Kappa score in the process), upgrading the model using GloVe, retraining the model, and adding more functionality in a Semantic Similarity and Grammatical/Spelling Accuracy finder. We used Django to create a website where you can submit essays to get them analyzed, as well as view past essays and their scores as well.
Challenges we ran into
Originally meant to be solely an SAT grader, we expanded it into the full range of essays since it was too difficult to train three separate AIs to look through the 3 categories of the SAT. Also, there wasn't nearly enough training data. We were going to use solely html, js, and css to create the website portion, but halfway through we realized it would be extremely difficult for us to implement the neural net on that. Instead we ended up having to learn Django from scratch in order to successfully implement our code!
Accomplishments that we're proud of
Successfully implementing Semantic Similarity Successfully implementing an AI model to score essays with relatively good accuracy Creating a website from Django, which we had little-no experience in prior
What's next for S. A. Grader
Making the model run faster both on its own (by changing the model to learn off GloVe) and in the browser. Beautifying the website further. Making it so that you can share essays and scores online. More accurate predictions as currently it's a bit harsh with grading.