Efficacy of Deep Neural Embeddings-Based Semantic Similarity in Automatic Essay Evaluation

Introduction

Automatic Essay Evaluation is one of the oldest research area in the field of Natural Language Processing (NLP). Unlike multiple choice questions and short question answers, an essay is an open ended question. There is no fixed format and one can have multiple ways of writing an essay. Manually grading the essays is a very resource intensive task from the perspective of time and labour. Teachers have to spend their valuable time on grading the essays written by the students. If we have an automatic essay grading system then teachers can devote more time on the teaching part. An essay is used to assess one's understanding of the particular language. Because of which, TOEFL (2019) and GRE (2019) like exams has essay writing as one of the main component. Since last 5 decades researchers are developing solutions for automatic essay grading systems (Page, 1968; Christie, 1999; Rudner et al., 2006). In Natural Language Processing field there has been many advancements in last couple of years. We have more powerful language models which can perform various tasks as par with humans (Young et al., 2018). In tasks like sentiment Analysis, Chatbot, Question Answering, Automatic Essay Evaluation, Dialogue Systems, Parsing, Word-sense disambiguation, Named-Entity Recognition, POS Tagging and many more, we are observing good results (Young et al., 2018; Khurana et al., 2017; Cambria & White, 2014). The computing resources are more available and affordable now, as compared with couple of years back. Due to this, the research in NLP using Deep Learning Techniques is taking new leap in every field (Otter et al., 2020; Young et al., 2018; Deng & Liu, 2018).