Application of the Support Vector Machine Algorithm for Sentiment Classification of Movin by Mandiri Taspen Application Reviews

Abstract

The mobile banking application Movin by Mandiri Taspen is a vital tool that supports digital financial and banking services for customers, particularly retirees and senior citizens. Evaluating user perception and satisfaction through Google Play Store reviews is essential for identifying technical bottlenecks and improving digital service quality. This study aims to analyze and classify user review sentiments for the Movin application using the Support Vector Machine (SVM) algorithm across three sentiment classes (Positive, Neutral, and Negative). A curated dataset of 12,303 review records was processed through text preprocessing stages (including cleaning, case folding, tokenizing, stopword removal, and stemming) and Term Frequency-Inverse Document Frequency (TF-IDF) feature weighting. The dataset was partitioned using an 80:20 ratio into a training set (9,842 samples) and a test set (2,461 samples). Hyperparameter optimization of the SVM algorithm yielded an optimal model configuration using a linear kernel with a regularization penalty parameter C = 10.0. Model evaluation showed strong performance, achieving an Accuracy of 95.98%, Macro Precision of 95.25%, Macro Recall of 94.42%, and a Macro F1-Score of 94.82%. Class-specific evaluations recorded F1-scores of 0.97 for the Neutral class, 0.96 for the Negative class, and 0.91 for the Positive class. The predominance of neutral and negative sentiments underscores the necessity for targeted enhancements in interface responsiveness and system stability. This study offers data-driven strategic insights for application developers to implement targeted technical refinements grounded in public opinion analysis.