A Robust Model for Automated Essay Scoring System

The primary challenges in essay assessment revolve around deciphering the semantic content embedded within the essays. Despite numerous AES systems, many still need to evaluate essays regarding domain-specific content and cohesion comprehensively. Consequently, it remains challenging to maintain the cohesive flow of text and ensure the preservation of sentence connectivity, which is crucial for capturing the holistic context. This research paper addressed these issues by evaluating essays based on semantics. We achieved this by implementing feature extraction techniques with a sentence encoder and a pre-trained transformer model. Additionally, we trained a bi-directional long short-term memory (Bi-LSTM) network to capture the coherence within essays effectively. Our model was tested using the Kaggle ASAP dataset, where it outperformed other established models, boasting an impressive average Quadratic Weighted Kappa (QWK) score of 0.812. We collected a domain-specific dataset related to operating system courses to ensure the model’s consistency across different datasets.