Deep Learning, NLU Aspect-Based Sentiment Analysis of Kimia Farma Pharmaceutical Services Using IndoBERT Deep Learning

Abstract

This study implements the Natural Language Understanding (NLU) architecture based on IndoBERT Base p2 to conduct Aspect-Based Sentiment Analysis on the review of the Kimia Farma Mobile application. Using 11,301 review data, the model was optimized through oversampling and Bayesian Optimization techniques to address data imbalances. The results of the study proved the success of the model with an accuracy of 97.51%, a significant jump compared to SVM-based basic research which only reached 85.10%. Technical performance is validated by a precision Recall value of 96.53%, precision of 98.28% and an f1-score of 97.62% on negative review data, demonstrating the model's acumen in detecting customer complaints. Key findings identified critical bottlenecks in OTP systems, real-time stock synchronization, and medication error risk. This research provides strategic recommendations for Kimia Farma's management to strengthen IT infrastructure and standardize digital clinical pharmacy education to ensure service quality.