Automated Essay Scoring Using Machine Learning

We built an automated essay scoring system to score approximately 13,000 essay from an online Machine Learning competition Kaggle.com. There are 8 different essay topics and as such, the essays were divided into 8 sets which differed significantly in their responses to the our features and evaluation. Our focus for this essay grading was the style of the essay, which is an extension on the studies conducted determining the quality of scientific articles by adding maturity to the feature set (Louis and Nenkova, 2013). An aspect of this project was to recognize the difference between the advanced nature of scientific articles to the coherency of middle to high school test essays. We evaluated Linear Regression, Regression Tree, Linear Discriminant Analysis, and Support Vector Machines on our features and discovered that Regression Trees achieved the best results with κ = 0.52. NOTE: I (Shihui Song) had mentioned that I was working with Jason Zhao from Machine Learning, but we divided the project into two tracks: one for Machine Learning and one for NLP. The NLP track resulted in this project but the Machine Learning track was focusing on other Machine Learning techniques that we did not get the chance to test on this project. I worked on the NLP aspect and the Machine Learning part as well, but as he’s not in the class, he was only consulting for the NLP project. This I would humbly request that you treat this as a single person project if possible.