Natural Language Processing for Automated Feedback Generation in Peer-Reviewed Online Assignments

Automated feedback generation enhances personalized learning by providing timely, context-aware feedback, fostering improved student performance and engagement. Prior approaches are mostly based on such finite-state models as rule-based systems or traditional machine learning, which do not have an adequate means of naturally incorporating deep patterns in text context. This decreases their ability to elicit well integrated and contextually relevant feedback. The proposed study addresses several shortcomings of current models such as limited language generation in feedbacks through the introduction of two methodologies, BERT for feature extraction and GRU for feedback. Using the contextualized word vectors of BERT and the sequential learning strength of GRU, the model provides relevant feedback on student’s writing assignments. By leveraging BERT in a custom set and focusing on parameters like number of layers, number of attention heads in each layer, and sequence length. These contextual embeddings are then passed through the GRU model in order to keep the feedback organized and coherent. The findings also illustrate enhancing BERT of GRU model to deliver a high accuracy of $94.17 \%$ compared to a past approach of precision and recall. The presented work proves the applicability of the advanced hybrid models for personalized educational feedback generation.