AUTO-GRADING ARABIC SHORT ANSWER QUESTION

Abstract

Automated Essay Grading Systems (AEGS) have become the main tools to address the challenges associated with manual essay grading, especially in Arabic. These systems use advanced NLP and Machine Learning techniques to support and enhance grading, rating efficiency, and equality. Despite significant improvements in automated grading systems, studies on Arabic essay evaluation are limited due to the Arabic language's unique morphological and syntactic complexities. This paper presents a unique automated grading system for short-form Arabic essay questions. The system applies text representation techniques, such as Word2Vec and TF-IDF, and text similarity techniques, such as LSA, LCS, Cosine Text Similarity, and Jaccard Text Similarity measurements. Furthermore, a stacking-based machine learning model supplies these estimates to attain a coherent and reliable grading system. The system's strength is determined by using metrics such as mean absolute error (MAE), Root Mean Square Error (RMSE), Pearson correlation, and Spearman correlation. Experimental results establish the validity of the unified technique in achieving high accuracy and strong correlations with human raters' ratings. The stacking model (TF-IDF + Jaccard + LCS + LSA) showed outstanding performance, resulting in negligible errors (MAE = 0.81, MSE = 0.96) and significant correlations (Pearson = 0.73, Spearman = 0.76). The advanced approach is very accurate and strongly correlates with human ratings, providing a scalable and economical solution for correcting Arabic text.