Big Data in Education: Assessment of the New Educational Standards

Currently, automatic essay scoring uses machine learning algorithms and models which are mainly trained on statistical text measures, such as word choice, text length, grammatical measures, and even the number of commas. The resulting models are not scalable, since they are highly domain-dependent and require large amounts of human-scored essay data. The model outcomes are scores, numbers that are correlated with human rater scores but that contain no information about essay content and provide almost no clues on the underlying reasons of this number’s value. We present a novel automatic essay scoring approach that evaluates essay content, does not require human-scored training data, is domain-independent, and provides feedback on student’s understanding or misconceptions. We show results of our algorithms on two small essay data sets–fourth and fifth grade students describing the hearing process and the vision process. Finally, we discuss the scalability and possible improvements and modifications of our current algorithms.