Analisis Sentimen pada Pemodelan Topik Ulasan Pengguna Gojek Menggunakan Metode Vader dan LDA

Abstract

A total of 319,031 user reviews of the Gojek application were collected from the Google Play Store between January 1, 2022, and December 31, 2024. The data underwent preprocessing stages including cleansing, normalization using five Indonesian slang dictionaries, translation into English, case folding, stopwords removal, tokenization, and lemmatization. Term weighting was performed using the TF-IDF method to generate more relevant word feature representations. Topic modeling with the Latent Dirichlet Allocation (LDA) algorithm identified nine main topics based on the highest coherence score of 0.563. These topics include: (1) Delivery Services, (2) Account & Login, (3) User Feedback, (4) Driver & Service Ethics, (5) User Satisfaction & Promotions, (6) Application Stability & Performance, (7) Transactions & Payments, (8) Ease & Efficiency of Service, and (9) User Needs & Activities. Sentiment analysis using the VADER method was conducted on each topic. The analysis results indicated that most user reviews expressed a positive sentiment, particularly in the User Feedback topic, which was dominated by 90.2% positive sentiment. Conversely, the Account & Login topic had the highest proportion of negative sentiment (17.0%) and neutral sentiment (64.5%). Sentiment classification was evaluated by comparing VADER results with manually labeled data, resulting in an accuracy of 79.12%, precision of 83.56%, recall of 79.12%, and an F1-score of 79.27%. These findings suggest that VADER is effective in analyzing sentiment in user reviews and can support application developers in gaining deeper insight into user perceptions and prioritizing service improvements.