Comparative Analysis of Sentiment Analysis Using Naive Bayes with TF-IDF, Bag of Word, and Word2Vec (Case Study: Lazada Electronic Product Reviews)

MUHAMMAD DZAKI P, Anny Kartika Sari, S.Si., M.Sc., Ph.D

2021 | Skripsi | S1 ILMU KOMPUTER

Electronic commerce (e-commerce) adalah platform online di mana penjual dan pembeli bertemu untuk bertukar produk atau layanan. Selain itu, perdagangan elektronik memainkan peran penting dalam pertumbuhan Usaha Kecil dan Menengah. Baru-baru ini, e-commerce menjadi populer yang mendorong bisnis untuk datang dengan strategi bisnis baru. Penjualan adalah metrik bisnis utama bagi para pemain e-commerce. Ulasan produk dalam e-commerce telah secara signifikan mengarahkan citra pedagang atau merek ke dalam penjualan. Menanggapi masalah ini, data pelanggan historis dalam e-commerce diharapkan dapat memberikan wawasan bisnis yang berharga melalui implementasi data mining, seperti text mining. Penelitian ini bertujuan untuk membandingkan model ekstraksi ciri pada Naive Bayes untuk masalah analisis sentimen. Analisis sentimen bermanfaat untuk pengambilan keputusan e-commerce, terutama untuk merek dan pedagang. Pendekatan penelitian menggunakan konsep text mining mulai dari pengumpulan data, pra-pemrosesan (tokenization, stop word removal, dan stemming), ekstraksi fitur, analisis sentimen, dan evaluasi. Dataset tersebut merupakan data e-commerce yang di-crawl dari produk elektronik Lazada dengan output matriks kebingungan ekstraksi fitur. Ekstraksi yang digunakan untuk penelitian ini adalah TF-IDF, Bag of Word, dan Word2Vec.

Electronic commerce (e-commerce) is an online platform where sellers and buyers meet to exchange products or services. In addition, electronic commerce plays an important role in Small and Medium Enterprise growth. Recently, e-commerce became popular which drives businesses to come up with fresh business strategies. Sales are the main business metric for e-commerce players. Product reviews in e-commerce have significantly led the merchant or brand image into sales. Responding to this issue, historical customer data in e-commerce is expected to give valuable business insight through data mining implementation, such as text mining. This research is aimed to compare feature extraction models on Naive Bayes for the sentiment analysis problem. The sentiment analysis is beneficial for e-commerce decision-making, especially for brands and merchants. The research approach is using text mining concepts starting from data gathering, pre-processing (tokenization, stop word removal, and stemming), feature extraction, sentiment analysis, and evaluation. The dataset is crawled e-commerce data from Lazada electronic products with the feature extraction confusion matrix as the output. The extractions used for this research are TF-IDF, Bag of Word, and Word2Vec.

Kata Kunci : Feature Extraction, Naive Bayes, Sentiment Analysis