Analisis Sentimen dan Pemodelan Topik Ulasan Aplikasi Noice Menggunakan XGBoost dan LDA

Abstract

In recent years, audio content has seen a rise in consumption. The COVID-19 pandemic also contributes to the rise in consumption. In the survey that was conducted in 2021, more than 40% of people in France, Germany, and Spain have been listening to more audio content since the first restriction on COVID-19 came into place [1]. One of the rising startups in Indonesia that offers audio content with their original and exclusive content is Noice. To maintain their quality of service, it’s important to look into the reviews that were written for their application. To analyze the reviews, sentiment analysis and topic modeling can be used to extract the sentiment polarity and the topics that are discussed on each sentiment polarity [3]. In this study, XGBoost and Latent Dirichlet Allocation are used to analyze the reviews that were written in Google Play Store. The result of the sentiment analysis yielded accuracy, precision, recall, and F1-score of 87,5%, 84,8%, 79,4%, and 81,6%. While the topic modeling managed to extract 16 and 6 topics respectively for positive and negative reviews.