Improved Automated Essay Grading System Via Natural Language Processing and Deep Learning

Abstract

In order to evaluate students’ total skills, educators repeatedly utilize questions based on free text to evaluate students’ total skills. Yet, when correction is done manually errors occur in addition to long time periods used, hard work, high costs and different opinions as to how to correct papers, a single paper is corrected by multiple people to avoid partiality. Thus, the smart system for automatic grading can solve the problem. Here, we present a very advanced system for grading essays automatically. This is based on Natural Language Processing and Deep Learning technologies. Thus, we need a system to automatically grade essays with low costs, less time and more accurate scores. We need thus an inelegant system for correcting essay questions on an automatic basis. We introduced a method which encodes essays in the form of sequential embeddings. We then use a long Short Term Memory Network (LSTM) working in two directions in order to register semantic information. This method also focuses concentration on each essay in order to be taught how to focus on those materials which are authentic in articles. We can also thus get a good proof of the result of prediction. This BI LSTM may be utilized also to produce neural networks which have the sequence information in the two directions: from the future to the past (backwards) or vice versa, which is called Bidirectional Long Short-Term Memory (BI-LSTM) (past to future). In order to train and test, we utilized the popular set of essays presented in the Automated Student Assessment Prize by Kaggle. The smart system used for automatic grading, in our research, predicts grades in an up-to-date manner. Moreover, the smart system for autograding we have proposed has the ability to highlight important words and sentences, evaluate the logical relationships in meaning in a sentence and gives us in advance grades that can be explained.