Natural Language Processing (NLP) for Sentiment Analysis of Service Management Reviews on Kimia Farma

Abstract

This study conducted sentiment analysis modeling on 11,301 reviews of the Kimia Farma Mobile application by integrating Natural Language Processing (NLP) techniques and sigmoid activation functions. Methodologically, this study evaluates the performance of Machine Learning algorithms in converting unstructured text into measurable sentiment categories. The test results showed that the Support Vector Machine (SVM) algorithm achieved the most optimal performance with an accuracy rate of 85.10%. Theoretically, the implementation of the sigmoid function has proven to be effective in mapping the probability of sentiment, which results in a decisive and systematic differentiation of categories. Empirical findings show a dominance of positive sentiment at 56.2%, reflecting the efficacy of digital pharmacy services on user satisfaction. However, the existence of neutral sentiment of 40.3% and negative sentiment of 3.5% identified a disparity between user expectations and system functionality, especially in terms of digital transaction stability and customer service responsiveness. As a development, it is recommended to use IndoBERT-based deep learning architecture and the Natural Language Understanding (NLU) approach to increase the depth of information extraction in the context of subjective and complex review languages.