Comprehensive Analysis of Teacher Teaching Performance Through Sentiment and POS Tagging

Abstract

This study aims to develop a sentiment analysis and POS (Part of Speech) system for student assessments of teachers in Indonesia. The system classifies assessment sentiment into positive and negative classes while identifying the topic of discussion. Testing results show that the w11wo/indonesian-roberta-base-sentiment-classifier model provides the best performance, achieving an accuracy of 0.87, with the highest precision, recall, and F1-Score for the positive class. The crypter70/IndoBERT-Sentiment-Analysis model ranks second with an accuracy of 0.84 but performs less optimally in detecting negative sentiment. Meanwhile, the mdhugol/indonesia-bert-sentiment classification model has the lowest performance with an accuracy of 0.80, particularly in predicting negative sentiment.