REKOMENDASI PRODUK E-COMMERCE BERBASIS KLASIFIKASI MENGGUNAKAN ALGORITMA MACHINE LEARNING

Abstract

The growth of e-commerce is driving an increase in the number of product reviews, but notall reviews are informative and relevant. This study aims to build a Tokopedia reviewclassification system using the Naive Bayes algorithm to filter relevant reviews. The datasetused totaled 40,607 product review lines. The preprocessing process includes cleaning,tokenization, stopword removal, and feature transformation using TF-IDF with 5,000 featurewords. Labels are determined based on ratings, where a rating of 4-5 is considered relevant.The Multinomial Naive Bayes model yielded an accuracy of 93.71%, a precision of 0.94, arecall of 0.99 for the relevant class, and an F1-score of 0.96. Although performance inirrelevant classes is still low, this model is effective in supporting product recommendations