Analyzing Multi-Label Sentiment in Educational Feedback: A Case Study in Higher Education

In the context of modern education, student feedback serves as a crucial source of data for improving teaching quality and shaping training strategies. However, most previous studies on student feedback analysis have been limited to assigning a single sentiment label to each sentence, without capturing the inherent multi-dimensionality of the data. This limitation leads to the omission of feedback that simultaneously expresses multiple sentiments. To address this gap, our study proposes a multi-label sentiment analysis approach that enables the identification of multiple emotional states within a single sentencelevel feedback. Specifically, we introduce ViSFD-SAUE-a novel dataset comprising 16,178 authentic feedback sentences collected from Vietnamese university students, manually annotated with an inter-annotator agreement of 94.08 %. Based on this dataset, we implement and compare the performance of various models, with results showing that PhoBERT achieves the best performance, reaching an F1-macro score of 0.8445. This study not only contributes the first Vietnamese multi-label sentiment analysis dataset in the educational domain but also opens new opportunities for applying AI to enhance the quality of higher education.