ANALISIS PERBANDINGAN KINERJA BERT DAN INDOBERT DALAM MENDETEKSI ULASAN PALSU PRODUK E-COMMERCE BERBAHASA INDONESIA

Marsha Kamila, NIM.: 22106050035 (2026) ANALISIS PERBANDINGAN KINERJA BERT DAN INDOBERT DALAM MENDETEKSI ULASAN PALSU PRODUK E-COMMERCE BERBAHASA INDONESIA. Skripsi thesis, UIN SUNAN KALIJAGA YOGYAKARTA.

	Text (ANALISIS PERBANDINGAN KINERJA BERT DAN INDOBERT DALAM MENDETEKSI ULASAN PALSU PRODUK E-COMMERCE BERBAHASA INDONESIA) 22106050035_BAB-I_IV-atau-V_DAFTAR-PUSTAKA.pdf - Published Version Download (5MB)
	Text (ANALISIS PERBANDINGAN KINERJA BERT DAN INDOBERT DALAM MENDETEKSI ULASAN PALSU PRODUK E-COMMERCE BERBAHASA INDONESIA) 22106050035_BAB-II_sampai_SEBELUM-BAB-TERAKHIR.pdf - Published Version Restricted to Registered users only Download (10MB) | Request a copy

Abstract

The rapid growth of e-commerce in Indonesia has increased the use of product reviews as an important consideration for consumers before making purchases. However, the emergence of fake reviews has become a serious issue because it can affect consumer trust and reduce the credibility of e-commerce platforms. This study aims to compare the performance of Multilingual BERT and IndoBERT models in detecting fake reviews written in Indonesian. This research employed an experimental method consisting of data collection, labeling, preprocessing, model training, evaluation, and result comparison. The dataset was obtained from Kaggle and consisted of 18,242 Indonesian e-commerce product reviews. Approximately 10% of the dataset was manually labeled and then used in a pseudo-labeling process with the Multinomial Naive Bayes algorithm to generate additional training data. Both Multilingual BERT and IndoBERT were trained using the same parameters to ensure a fair comparison. Model evaluation was conducted using accuracy, precision, recall, F1-score, and confusion matrix metrics. The results showed that IndoBERT achieved the best performance with an accuracy of 87.12%, precision of 85.81%, recall of 83.01%, and F1-score of 84.39%. Meanwhile, Multilingual BERT achieved an accuracy of 86.30%, precision of 83.23%, recall of 84.31%, and F1-score of 83.77%. These findings indicate that IndoBERT is more effective in understanding the characteristics of the Indonesian language, especially informal e-commerce review texts. This study also demonstrates that the pseudo-labeling approach can help increase the amount of training data, although the quality of additional labels greatly affects model performance. Based on the research findings, IndoBERT is more recommended for implementing fake review detection systems on Indonesian e-commerce platforms compared to Multilingual BERT.

Share this knowledge with your friends :

Actions (login required)

View Item