EVALUASI PERFORMA METODE LONG SHORT TERM MEMORY (LSTM) DAN RECURRENT NEURAL NETWORK (RNN) PADA ANALISIS SENTIMEN KOMENTAR PENGGUNA APLIKASI KITALULUS

Abstract

Sentiment analysis is one of the essential techniques in natural language processing for identifying user opinions toward a product or service. This study aims to evaluate the performance of the Long Short Term Memory (LSTM) and Recurrent Neural Network (RNN) methods in sentiment analysis of user comments on the Kitalulus application. A total of 2,500 comments collected from user reviews were utilized and underwent several preprocessing stages, including case folding, tokenization, stopword removal, stemming, and padding. The processed data were then trained using both RNN and LSTM models with similar architectural configurations. The experimental results show that the LSTM method outperforms RNN, achieving the highest accuracy of 91.51%, while RNN attained 88.48%. These findings demonstrate that LSTM is more effective in capturing long-term dependencies in textual data, making it more suitable for sentiment analysis of user comments on the application.