Sentiment Analysis on the Shopee E-Commerce Application using the Support Vec-tor Machine (SVM) Algorithm (A Case Study of Play Store User Reviews)
Abstract
The rapid development of e-commerce platforms makes user reviews on the Google Play Store a crucial metric for evaluating customer satisfaction, as seen in the Shopee application which currently dominates the Indonesian digital market. The high volume of unstructured review data requires an automated computational approach to extract the information within it. This study aims to analyze the sentiment of Shopee application user reviews using the Support Vector Machine (SVM) algorithm guided by the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology. Data collection was carried out through web scraping techniques totaling 10,000 raw reviews, which after going through the text pre-processing (NLP) stage left 8,095 clean data. Text feature extraction used the Term Frequency-Inverse Document Frequency (TF-IDF) weighting method. With a train and test data split scheme of 80:20 and the application of a Linear kernel, the SVM model proved reliable by producing an overall accuracy of 85.30%. The model performed optimally in the majority class with the positive sentiment F1-Score reaching 90.60% and negative 81.15%. However, the model failed to predict neutral sentiment due to an extreme dataset imbalance (4.77%). Word Cloud and Bigram visualizations revealed that positive sentiment was driven by the application interface and "free shipping" promos, while negative sentiment purely originated from logistics supply chain and shipping service constraints. This study proves the reliability of the SVM algorithm in high-dimensional text processing and recommends the application of data balancing techniques (such as SMOTE) for future research, as well as suggesting Shopee's management to prioritize the evaluation of expedition partners.