The idea of prioritizing customer satisfaction to uphold or improve the excellence of a product or service can incorporate the utilization of user feedback. However, to provide a comprehensive visual summary to application developers or stakeholders, it is important to provide a detailed description of user sentiment issues. In this study, the data source used is user feedback from local telemedicine applications in Indonesia. This research builds a framework of deep learning to perform user feedback analysis and applies topic modeling to sentiment clusters. then builds visual construction of research results effectively and efficiently, to facilitate stakeholders in making decisions. Build a framework to analyze user feedback utilizing deep learning BiLSTM + IndoBERT for sentiment classification and LDA to model topics in sentiment groups. The results show that most of the user reviews of the five telemedicine applications have a positive sentiment at 91%. The model used has good prediction performance with the accuracy of the BiLSTM model with IndoBERT 96.44%. The negative sentiment group comprises 12 topics (0.58446), the most significant topics being 35.4% about telephone broadcasting, 25.3% payments, and 8.5% about medicine purchase service. For the imbalanced data case, BiLSTM showed good accuracy and precision values. The classifications and topics generated by deep learning models are affected by improper data labeling, so it is necessary to explore the data labels generated.
Copyrights © 2024