Comparative Analysis of Logistic Regression and Support Vector Machine for Sentiment Classification of Shopee App Reviews on the Play Store

Abstract

The rapid advancement of information technology has significantly transformed consumer shopping behavior on mobile e-commerce platforms, with Shopee standing out as one of the most popular choices. User reviews provide essential insights into customer sentiment and preferences, thereby highlighting the importance of automated sentiment analysis for capturing public opinion. This study aims to evaluate and compare the performance of two supervised learning algorithms Support Vector Machine (SVM) and Logistic Regression (LR) in classifying sentiments from 3,000 Shopee app reviews obtained from the Playstore. The research employed a 5-fold cross-validation method, with performance measured using accuracy, precision, recall, and F1-score. The findings indicate a clear trade-off between the models: SVM achieved a slightly higher test accuracy of 80.67% compared to 80.33% for Logistic Regression, but it showed greater susceptibility to overfitting. Conversely, Logistic Regression demonstrated more stable performance, maintaining a balanced trade-off between precision and recall without noticeable overfitting. This trade-off is likely due to SVM's complexity and sensitivity to high-dimensional feature spaces, whereas Logistic Regression's simpler linear model offers greater generalizability on this dataset. Based on these results, Logistic Regression is recommended for applications that prioritize consistency and robustness, while SVM may be preferable in contexts where maximizing precision is the primary objective.