Fake Listing or Truth? Using Pre-trained Deep Learning Model with Data Augmentation to Detect the Imposter

Abstract

In 2020, after the rising value of cryptocurrency, Graphics Processing Unit (GPUs) became shortage due to many scalper. To combat with this issue, there were some eBay users that trick the scalper bot with fake description or image in the listing. In this articles, we compare baseline machine learning models (Multinomial Naïve Bayes from Tf-idf vector, Logistic Regression, Support vector machine, Gradient Boosting classifier and XGBoost classifier) with deep learning models (Resnet-34 and Resnet-50 for image classifier, BERT and FLAIR-model for text classification) in order to detect these listings. As the data was imbalance, we used data augmentation to enhance the number of fake listing class for both images and text description. All models can achieve accuracy up to 90% except Logistic Regression. XGBoost and BERT are the best accuracy models when using with data augmentation. The accuracy are over 98% for both models.