Natūralios kalbos apdorojimo ir mašininio mokymosi metodų taikymas prognozuojant rašinių įvertinimus / ; Applying natural language processing and machine learning techniques to predict essay grades.

The aim of this study is to use different natural language processing and machine learning techniques to predict grades of student written essays. During this study, two methods of creating sets of features were used: feature engineering on raw text and word embedding extraction from Transformer-based model. These feature sets then were used to train Support Vector Regression model. The results show that for all types of essay grades, more accurate predictions in terms of RMSE and MAE metrics are obtained while using word embeddings, but the differences in the magnitude of the errors are negligible. During the course of this study, it was observed that extraction of embeddings is much more computationally expensive than using simple feature engineering methods, and, therefore, it cannot be confidently stated that this method is more suitable.