ASPECT-BASED SENTIMENT ANALYSIS ON STUDENT’S FEEDBACK IN VIETNAMESE

Abstract

In recent years, universities are interested in surveying and analyzing student’s feedbacks to improve teaching effectiveness as well as training quality. However, the manual analysis will be costly in terms of effort and time-consuming with the large data. Therefore, in this paper, we introduce a new dataset on student’s feedback of aspect categories detection and aspect-sentiment classification tasks. Our data consists of 5,010 sentences which are annotated by 11 pre-defined aspect categories (teacher behavior, teaching skills…) and 3 sentiment polarities (positive, negative, neutral) with annotation agreements of 88.95% and 80.52% according to two tasks. In addition, we present a series of experiments on the dataset based on a combination model BiLSTM-CNN, compared with other machine learning approaches. The experimental results show that our combination method achieves the best scores with the F1-score of 78.93% and 73.78% for the aspect category detection task and aspect-sentiment classification task, respectively. Experimental results demonstrate the effectiveness of our ensemble architecture.