Abraham Adrian Yanuar Laik, Adinda Nabilla, Andi Diah, Sumanto, Ahmad Indra, Yudi Arya

Komparasi Naive Bayes dan SVM untuk Analisis Sentimen Pada E-Commerce Seller Center

Introduction

Komparasi naive bayes dan svm untuk analisis sentimen pada e-commerce seller center. Komparasi Naive Bayes & SVM untuk analisis sentimen ulasan pelanggan e-commerce Tokopedia & TikTok Shop. Naive Bayes unggul dengan akurasi 97.50% & F1-score 84.00%.

Abstract

The development of e-commerce drives the need to understand customer opinions through sentiment analysis to improveservice quality. Tokopedia and TikTok Shop as popular e-commerce platforms provide a review feature that can be asource of data to analyze consumer perceptions. This study aims to compare the performance of two text classificationalgorithms, namely Naive Bayes and Support Vector Machine (SVM), in analyzing the sentiment of customer reviews takenfrom the TikTok Tokopedia Seller Center dataset. The research method used is a computational experiment with aquantitative approach. The dataset used is sourced from the Kaggle site and is available in clean and labeled conditions(positive and negative). Model evaluation is done by measuring accuracy, precision, recall and F1-score. The results showthat Naive Bayes is superior with 97.50% accuracy and 84.00% F1-score, compared to SVM which obtained 94.90%accuracy and 76.80% F1-score. Thus, Naive Bayes is considered more effective for sentiment analysis of e-commercecustomer reviews

Review

This study presents a comparative analysis of Naive Bayes and Support Vector Machine (SVM) algorithms for sentiment analysis on e-commerce customer reviews, specifically focusing on data from Tokopedia and TikTok Shop Seller Centers. The research addresses a relevant and timely problem given the increasing importance of understanding customer feedback for service improvement in the burgeoning e-commerce sector. The objective of comparing two well-established text classification algorithms using a readily available dataset is clear and contributes to the practical understanding of their applicability in this specific domain. The paper clearly articulates its goal of evaluating model performance through standard metrics, setting a straightforward and measurable research agenda. The methodology employed is a computational experiment with a quantitative approach, which is appropriate for algorithm performance comparison. The choice of Naive Bayes and SVM as benchmark algorithms is sound, representing both probabilistic and margin-based classification paradigms. The use of a Kaggle dataset, described as clean and labeled (positive and negative), simplifies the experimental setup and allows for a direct comparison of the algorithms' intrinsic performance without the added complexity of data preprocessing. The evaluation metrics—accuracy, precision, recall, and F1-score—are standard in text classification and provide a comprehensive view of model performance. The findings, indicating Naive Bayes' superiority with 97.50% accuracy and 84.00% F1-score over SVM's 94.90% accuracy and 76.80% F1-score, are clearly presented and support the conclusion that Naive Bayes is more effective for this specific task and dataset. While the study provides a valuable comparison, a deeper discussion on the potential reasons behind Naive Bayes' superior performance in this context, possibly related to the characteristics of e-commerce review text (e.g., strong keyword indicators for sentiment), would enhance the theoretical contribution. Additionally, exploring the impact of various text representation techniques (e.g., TF-IDF variations, word embeddings) beyond what's implied by "clean and labeled conditions" could offer further insights into optimizing sentiment analysis for this domain. Future work could also consider the generalizability of these findings to other languages or more complex, multi-class sentiment problems. Nevertheless, this paper offers a solid and practical contribution to the application of machine learning in e-commerce sentiment analysis, providing a clear benchmark for practitioners and researchers.

Full Text

Comments

You need to be logged in to post a comment.