Optimasi Analisis Sentimen Ulasan Sunscreen di E-Commerce Menggunakan Algoritma SVM dan SMOTE

Abstract

Abstrak : 

Analisis sentimen terhadap ulasan pengguna di e-commerce membantu produsen memahami kepuasan pelanggan. Penelitian ini bertujuan untuk menganalisis sentimen ulasan produk sunscreen di Facetology Official Shop menggunakan algoritma Support Vector Machine (SVM). Data ulasan dikumpulkan melalui scraping, diberi label secara manual, dan diproses menggunakan metode preprocessing seperti data cleaning, Case Folding, tokenizing, stopword removal, serta SMOTE untuk menyeimbangkan data. Ekstraksi fitur dilakukan dengan TF-IDF, dan SVM digunakan untuk mengklasifikasikan sentimen menjadi positif, negatif, dan netral. Hasil penelitian menunjukkan model SVM dengan kernel linear mencapai akurasi 93%, presisi keseluruhan 95%, recall 91%, dan F1-Score 93%. Pendekatan ini menunjukkan peningkatan performa model dengan akurasi 93% setelah penerapan SMOTE untuk penyeimbangan data. Sentimen mayoritas positif, mengindikasikan tingkat kepuasan tinggi, meskipun ada ulasan negatif terkait efek samping produk. Teknik preprocessing dan penyeimbangan data terbukti efektif dalam meningkatkan performa model. Pendekatan dapat diaplikasikan untuk analisis sentimen produk serupa guna mendukung pemahaman perusahaan terhadap konsumen

==================================================

Abstract :

Sentiment analysis of user reviews on e-commerce platforms helps producers understand customer satisfaction. This study aims to analyze the sentiment of sunscreen product reviews in the Facetology Official Shop using the Support Vector Machine (SVM) algorithm. Review data were collected through scraping, manually labeled, and processed using preprocessing methods such as data cleaning, case folding, tokenizing, stopword removal, and SMOTE to balance the data. Feature extraction was performed using TF-IDF, and SVM was used to classify sentiments into positive, negative, and neutral categories. The results show that the SVM model with a linear kernel achieved an accuracy of 93%, an overall precision of 95%, a recall of 91%, and an F1-Score of 93%. This approach demonstrated improved model performance, with 93% accuracy achieved after applying SMOTE for data balancing. The majority of sentiments were positive, indicating a high level of customer satisfaction, although some negative reviews mentioned side effects of the product. The preprocessing techniques and data balancing proved effective in enhancing the model's performance. This approach can be applied to sentiment analysis of similar products to support companies in better understanding their consumers.