ARTIFICIAL INTELLIGENCE IN EDUCATION: ANSWER SCORING USING WORD EMBEDDING & DEEP LEARNING - AN LSTM NETWORK ANALYSIS

Abstract

The increasing number of students in education has prompted the exploration of Automated Answer Scoring (AES) as a way to efficiently handle educational tasks and the goal is to assign grades to essays and provide feedback using computer. The objective of this project is to create deep learning models designed to automate the process of essay scoring, followed by a thorough evaluation of their performance.In this research, a kaggle repository containing student's essay responses and their corresponding gradings recorded data set was used to train and test the efficacy of the adopted techniques. Pre-processing by splitting essays into words, Inputting manually crafted features and helping to convert them into vector format using woodstoves technique, Natural language processing(NLP) techniques was used to extract features from essays in the dataset. The chosen dataset underwent analysis with LSTM network algorithms, and it was split into two distinct segments: training data and testing data. The assessment involved comparing the inter-rater reliability and performance of these models against each other and against human graders. Among the various machine learning and conventional deep learning(DL) models examined, the LSTM network exhibited the highest level of agreement with human scorers, as evidenced by its achievement of the lowest mean absolute error for the test dataset.