Aspect-based sentiment analysis on student reviews using the Indo-Bert base model

Abstract

This study aims to gain a deeper understanding of online student reviews regarding the learning process at a private university in Indonesia and to compare the effectiveness of several algorithms: Naive Bayes, K-NN, Decision Tree, and Indo-Bert. Traditional Sentiment Analysis methods can only analyze sentences as a whole, prompting this research to develop an Aspect-Based Sentiment Analysis (ABSA) approach, which includes aspect extraction and sentiment classification. However, ABSA has inconsistencies in aspect detection and sentiment classification. To address this, we propose the BERT method using the pre-trained Indo-Bert model, currently the best NLP model for the Indonesian language. This study also fine-tunes hyperparameters to optimize results. The dataset comprises 10,000 student reviews obtained from online questionnaires. Experimental results show that the aspect extraction model has an accuracy of 0.890 and an F1-Score of 0.897, while the sentiment classification model has an accuracy of 0.879 and an F1-Score of 0.882. These results demonstrate the effectiveness of the proposed method in identifying aspects and sentiments in student reviews and provide a comparison between the four algorithms.