Analisa Perbandingan Latent Semantic Indexing (LSI) dan Latent Dirichlet Allocation (LDA) untuk Topic Modelling Aplikasi Identitas Kependudukan Digital (IKD)

Abstract

This study aims to analyze and compare two topic modeling methods, Latent Semantic Indexing (LSI) and Latent Dirichlet Allocation (LDA), in understanding user reviews of the Digital Population Identity (IKD) Application obtained from the Google Play Store. The main problem addressed is the large number of user reviews with diverse topics that are difficult to categorize manually, necessitating an automated method to identify the main themes in the data. The research process began with scraping 5,000 recent reviews, followed by data preprocessing (Remove Punctuation, Lowercase, and Tokenization) and vectorization using Bag of Words and DOC2BOW. Subsequently, topic modeling was performed using LSI and LDA, and the results were evaluated using the Coherence Score metric. The findings indicated that Latent Dirichlet Allocation (LDA) outperformed LSI, achieving a Coherence Score of 0.4163 compared to LSI's 0.3512, indicating that Latent Dirichlet Allocation (LDA) is more effective in identifying hidden topics within user reviews. Latent Dirichlet Allocation (LDA) is a superior method for topic modeling in IKD application reviews and can assist developers in understanding user needs and issues, thereby enhancing the application's service quality.