A GCN- and Deep Biaffine Attention-Based Classification Model for Course Review Sentiment

Abstract

In recent years, the increasing use of online surveys for course evaluation in schools has led to an outpouring of evaluation texts. These texts, with their emotional polarity, can give schools the most direct feedback. Emotional analysis on course evaluation, therefore, has great implications. However, the not-so-rigid text grammar and rich text content pose a challenge for sentiment analysis in Chinese course evaluation. To solve this problem, this paper proposes a sentiment classification model BiLSTM-GCN-Att (BGAN). Here, BiLSTM is used to extract the features of the text and output the hidden state vector. Then, the deep biaffine attention mechanism is used to analyze the dependence of the text and generate a dependency matrix. Next, input the hidden state vector to the GCN. Finally, the softmax function is used as the output layer of the model to perform sentiment classification. The model proves effective and experimental results, showing that the BGAN achieved a maximum improvement of 11.02% and 14.47% in precision and F1-score respectively compared with the classical models.