AutoGrading Classification of Tamil Descriptive Answers FastText and Roberta-Base Sentence Embedding

Considering the challenges associated with manually grading student answer scripts and the significance of the language, an automated approach is developed for assessing Tamil descriptive answers. Three-way assessment system of correct, incorrect, and partially correct is used to improve the autograding categorization accuracy. A total of eight datasets derived from four different dataset's sentence embedding produced by sophisticated natural language processing techniques RoBERTa-Base and FastText was created. To determine the optimal combination of data input, ten models inclusive of deep learning and machine learning models were utilized to perform a grade-based classification. This classification study proved that combination of the QuestionAnswer dataset with FastText embedding yielded the most promising result of F1 Score as 92.1%. The automatic grading methods tailored to the nuances of Tamil language are improved by this research.