PERANCANGAN ALAT BANTU ANALISIS SENTIMEN BERDASARKAN ULASAN PENGGUNA APLIKASI POST DARI GOOGLE PLAY STORE UNTUK PENDUKUNG KEPUTUSAN

• Main Author: Rayhan Fuadi, M.
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/76224
• Institution: Institut Teknologi Bandung

POST App is an information system application of transaction and reporting for various types of businesses. Currently, when compared to similar cash register applications, POST App has a low user satisfaction rating and number of downloads on Google Play Store. Therefore, the developers of the POST App need an effort to improve the app's performance from user reviews through user sentiment information and several dominant topics on why users like or dislike the POST App. This knowledge becomes an input for the company to support operational or managerial decisions and to plan strategic business steps to improve the competitiveness of the application. In this study, the CRISP-DM methodology is used with a text mining and natural language processing (NLP) approach to perform sentiment analysis and topic modeling of POST App user reviews from Google Play Store. The sentiment analysis model was not built independently, but uses the pre-trained IndoBERT- Classification model with 95% accuracy on test data. As for the topic model, the Latent Dirichlet Allocation model is used. The results of this modeling are then used to create a web application prototype with the Streamlit framework and made the knowledge base for the Generative Pretrained Transformers (GPT) model from OpenAI as a tool to assist in the extraction of sentiment analysis for the product management, data analysis, and customer service of the POST App. System validation results show that the prototype system has good functionality and can be applied in the company and has potential for further development..