<![CDATA[The rapid growth of digital technology has driven innovation in the beauty industry, one of which is the Soco by Sociolla platform that provides online product reviews. The increasing number of user reviews offers opportunities for conducting sentiment analysis to understand users’ perceptions of service quality. The main challenge in modeling sentiment for beauty product reviews lies in the use of highly varied, subjective, and informal language, which results in diverse distribution patterns. Therefore, this study not only applies the Support Vector Machine (SVM) algorithm for sentiment classification but also compares two kernels—Linear and Radial Basis Function (RBF)—and evaluates the effect of hyperparameter optimization using Grid Search in the context of beauty e-commerce data. A total of 3,387 reviews were collected and processed through several stages, including text preprocessing, labeling, feature extraction using TF-IDF, data splitting, model training, and evaluation. The results show that the baseline RBF kernel provides the best performance with an accuracy of 88.5%, while the baseline Linear kernel achieves an accuracy of 87.76%. Meanwhile, Grid Search optimization produces an accuracy of 86.22%, indicating that the explored hyperparameter configurations were unable to exceed the performance of the RBF baseline despite delivering stable results during cross-validation. These findings suggest that the linguistic characteristics of beauty reviews are more effectively addressed by non-linear kernels, making them superior to Linear kernels in recognizing non-linear patterns within user review data. Furthermore, the results indicate that hyperparameter optimization does not always lead to increased model accuracy, particularly when the baseline SVM configuration is already performing near optimally for the characteristics of the dataset used.]]>
Copyrights © 2025
