An AES System to Assist Teachers in Grading Language Proficiency and Domain Accuracy Using LSTM Networks

Abstract

Automated Essay Scoring (AES) is a task of automatically grading the students' answers to subjective or essay type questions. AES is an area where assessing the answers rationally is very important. Assessing these subjective answers has always been a challenging process concerning reliability and effort. In such times, where the entire education system has shifted to being online, it becomes necessary to develop a system that assesses students based on their subjective answers. However, the existing AES system primarily focuses on assessing essays on a single dimension that is either grading domain accuracy or grading the language correctness of the answers. Moreover, there are few AES systems to grade student’s responses in the computer science domain. To address these gaps, we propose an AES system to grade the subjective answers of students from the computer science domain. The proposed system grades the student’s responses in two dimensions, namely domain accuracy, and language proficiency. In order to test the system, we collected data from 200 students and manually labeled them for domain accuracy and language proficiency. The system graded the student’s responses automatically with domain accuracy of 89.47 percent and language proficiency of 84.79 percent.