Sentiment Analysis of Student Feedback for Learning Improvement using Naive Bayes and LSTM (Long Short-Term Memory) Methods

Abstract

This study aims to analyze the sentiment of STMIK Kaputama students’ comments using Lexicon-Based Labeling, Multinomial Naive Bayes, and Long Short-Term Memory (LSTM) approaches. The data consisted of student comments related to services, facilities, and learning processes within the campus environment. The research stages included automatic sentiment labeling using an Indonesian sentiment lexicon, text preprocessing (case folding, cleansing, tokenization, padding, and truncating), splitting the data into training and testing sets, training the Naive Bayes model using TF-IDF feature representation, and training the LSTM model using tokenized sequence data. The results show that both models were able to perform sentiment classification effectively on the available testing data. Although the obtained evaluation scores were very high, the results were still influenced by the limited number of data samples and the imbalance of class distribution. Conceptually, LSTM has an advantage in understanding context and word order, while Naive Bayes is simpler and more efficient for small-sized datasets. This study demonstrates that the combination of lexicon-based labeling and sentiment classification methods can be used as an initial approach to support the evaluation of service quality and student satisfaction at STMIK Kaputama.

This study aims to analyze the sentiment of STMIK Kaputama students’ comments using Lexicon-Based Labeling, Multinomial Naive Bayes, and Long Short-Term Memory (LSTM) approaches. The data consisted of student comments related to services, facilities, and learning processes within the campus environment. The research stages included automatic sentiment labeling using an Indonesian sentiment lexicon, text preprocessing (case folding, cleansing, tokenization, padding, and truncating), splitting the data into training and testing sets, training the Naive Bayes model using TF-IDF feature representation, and training the LSTM model using tokenized sequence data. The results show that both models were able to perform sentiment classification effectively on the available testing data. Although the obtained evaluation scores were very high, the results were still influenced by the limited number of data samples and the imbalance of class distribution. Conceptually, LSTM has an advantage in understanding context and word order, while Naive Bayes is simpler and more efficient for small-sized datasets. This study demonstrates that the combination of lexicon-based labeling and sentiment classification methods can be used as an initial approach to support the evaluation of service quality and student satisfaction at STMIK Kaputama.