Exploring Deep Reinforcement Learning and BigBird-BiLSTM Models for Automated Essay Grading

Salma Elhag, Abeer Alhattami

Abstract: This paper evaluates the potential of Deep Reinforcement Learning (DRL) and BigBird-BiLSTM models in enhancing Automated Essay Grading (AEG) systems. Leveraging the Hewlett dataset, the study examines how these models handle semantic features and scalability challenges compared to existing frameworks. Evaluation metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and R-squared (R2) highlight the strengths and limitations of each model.

Keywords: Automated essay grading, deep reinforcement learning, BigBird-BiLSTM, semantic features, evaluation metrics