Efektivitas IndoBERT pada Klasifikasi Sentimen Evaluasi Dosen: Studi Komparatif Support Vector Machine dan Naive Bayes

Abstract

Sentiment analysis of student feedback plays an important role in evaluating the quality of teach-ing and learning processes in higher education. Qualitative comments in Student Evaluation of Teaching (SET) provide deeper insights than numerical ratings. However, they are expressed in unstructured textual form, making large-scale analysis difficult to conduct consistently and sys-tematically. Therefore, Natural Language Processing (NLP) approaches are required to automati-cally identify sentiment tendencies within student comments. This study aims to compare the per-formance of Gaussian Naive Bayes and Support Vector Machine (SVM) algorithms for classifying sentiment in SET comments using IndoBERT-based text embeddings. The research follows the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework, including stages of data understanding, data preparation, modeling, evaluation, and deployment. Text comments were preprocessed and transformed into numerical vectors using IndoBERT Sentence-BERT em-beddings to capture contextual semantic relationships between words. These embeddings were then used as input features for both classification algorithms. Evaluation results show that the In-doBERT + SVM model achieved an accuracy of 93.88%, outperforming IndoBERT + Naive Bayes which obtained 92.40%. The SVM model also demonstrated more balanced precision, recall, and F1-score values across sentiment classes. These findings indicate that SVM is more effective in uti-lizing high-dimensional contextual embeddings for sentiment classification of student feedback.