A Hybrid PhoBERT-CNN-LSTM Model for Sentiment Analysis of Vietnamese Student Feedback

Abstract

Student feedback plays a crucial role in improving educational quality and creating an effective learning environment. This study applies sentiment analysis to Vietnamese student feedback to extract and classify emotions into positive, negative, and neutral categories, providing valuable insights to support teaching improvements. We propose a method that utilizes the PhoBERT model for semantic feature extraction, followed by a CNN-LSTM architecture to capture both local features and sequential relationships in feedback data. Experimental results on the UIT-VSFC dataset demonstrate that the proposed PhoBERT-based CNN-LSTM model achieves an accuracy of 93.24% and an F1-score of 92.92%. This model demonstrates superior performance compared to several other advanced approaches. It surpasses ensemble model, which achieved an F1-score of 92.79%. These findings confirm the effectiveness of the model in extracting and classifying sentiments from student feedback while proposing a practical approach for analyzing Vietnamese educational data, contributing to teaching quality enhancement.