Analisis Sentimen Ulasan Pengguna Aplikasi ChatGPT pada Google Play Store Menggunakan IndoBERT dan Support Vector Machine

Abstract

The rapid advancement of Large Language Models (LLMs) has significantly increased the adoption of ChatGPT for various activities, including learning, writing, programming, and information retrieval. The growing number of users has generated a substantial volume of reviews on the Google Play Store, reflecting user experiences, satisfaction levels, and various issues encountered while using the application. This study aimed to analyze the sentiment of ChatGPT application user reviews on the Google Play Store. The method employed was a hybrid approach combining IndoBERT and Support Vector Machine (SVM) on 312 manually labeled and pre-processed reviews. Data pre-processing (n=312) involved _case folding_, _tokenizing_, and _stop word removal_, followed by manual labeling which revealed a sentiment distribution of 64.10% positive, 24.04% negative, and 11.86% neutral. The hybrid IndoBERT-SVM model demonstrated exceptional sentiment classification performance, achieving an accuracy of 93.55%, precision of 92.80%, _recall_ of 93.55%, and an F1-score of 93.17%. This performance significantly surpassed other classification methods; IndoBERT-SVM's accuracy was higher compared to Naïve Bayes (87.00%) and a standalone SVM (92.00%) in comparative studies. This superiority stems from IndoBERT's ability to extract rich semantic and contextual feature representations from Indonesian text, which is then optimized by SVM to find an optimal classification _hyperplane_. This study extends the understanding of applying transformer-based pre-trained language models in multiclass sentiment analysis within the Indonesian language context, showcasing the superiority of the hybrid approach.