Identifying Fraud Sellers in E-Commerce Platform

Identifying fake reviews in e-commerce is crucial as they might impact buyers' purchasing decisions and overall satisfaction. This work investigates the effectiveness of machine learning and transformer-based models for detecting fake reviews on the Amazon Fake Review Labelled Dataset. The dataset contains 20,000 computer-generated and 20,000 original reviews across various product categories with no missing value. In this study, machine learning and transformer-based models were compared, revealing that transformer-based models outperformed in detecting fake reviews, achieving an accuracy of 98% with the DistilBERT model. Additionally, this work too examines the impact of word embedding on machine learning models in enhancing fake review detection accuracy. The results show that the word embedding model Word2Vec displays notable improvements, achieving accuracies of 92% with SVM and 90% with Random Forest and Logistic Regression. Furthermore, a comparison study was carried out on comparing transformer models from previous work, which utilized the same full dataset; it was found that the DistilBERT model produced comparable accuracy despite its lighter architecture. In summary, this study underscores the effectiveness of transformer-based models and machine learning models in detecting fake reviews while at the same time highlighting the importance of word embedding techniques in enhancing the performance of machine learning models. This work is hoped to contribute to combating fake reviews and fostering trust in e-commerce platforms.