Sentiment Analysis And Topic Modeling on User Reviews of Online Tutoring Applications Using Support Vector Machine and Latent Dirichlet Allocation

Abstract

<p>Ruangguru is an online non-formal education application in Indonesia. There are several appealing features that encourage students to study online. The app's release on the Google Play Store will assist app developers in receiving feedback through the review feature.Users submit various topics and comments about Ruangguru in the review feature of Ruangguru, making it difficult to manually identify public sentiments and topics of conversation. Opinions submitted by users on the review feature are interesting to research further. This study aims to classify user opinions into positive and negative classes and model topics in both classes. Topic modeling aims to find out the topics that are often discussed in each class. The stages of this study include data collection, data cleaning, data transformation, and data classification with the Support Vector Machine method and the Latent Dirichlet Allocation method for topic modeling. The results of topic modeling with the LDA method in each positive and negative class can be seen from the coherence value. Namely, the higher the coherence value of a topic, the easier the topic is interpreted by humans. The testing process in this study used Confusion Matrix and ROUGE. The results of model performance testing using the Confusion Matrix are shown with accuracy, precision, recall, and f-measure values of 0.9, 0.9, 0.9, and 0.89, respectively. The results of model performance testing using ROUGE resulted in the highest recall, precision, and f-measure of 1, 0.84, and 0.91. The highest coherence value is found in the 20th topic, with a value of 0.318. Using the Support Vector Machine and Latent Dirichlet Allocation algorithms are considered adequate for sentiment analysis and topic modeling for the Ruangguru application.</p>