Combination between Deep Learning and Transformer Models to Detect Fake Yelp Electronic Product Reviews

Abstract

Online product reviews play a major role in the success or failure of an E-commerce business. Before procuring products or services, the shoppers usually go through the online reviews posted by previous customers to get recommendations of the details of products and make purchasing decisions. There has been a rise in illusive review spam, which are fake reviews that are designed to appear genuine. Fake, strident, spam, misleading reviews are those written by those who do not have personal experiences with the topics of the reviews. Spammers spread fake reviews in order to denigrate or promote a specific brand or product, persuading consumers to purchase from that brand or not. The detection of genuine ratings and ratings-based reviews across the entire online platform, particularly Yelp product datasets, is the secondary objective. For the purpose of identifying fake online reviews in the e-commerce industry, the paper makes a number of novel hybrid techniques (transformer-based & deep learning), including BiLSTM-CNN, BERT-CNN and RoBERTa-CNN. According to the trial results, the BiLSTM-CNN procedure productively identifies counterfeit internet based audits with a high accuracy of 90% whereas other hybrid models also perform competitively. Moreover, it BiLSTM-CNN model exhibits the most favorable combination of training and testing times among the evaluated models.