A Research on Cross-Language Fake Reviews Identification Based on ERNIE and SGAN

Abstract

With the rise of social media, the prevalence of fake reviews has surged, causing significant harm to the e-commerce industry's competitive landscape. To tackle the issue of limited publicly available datasets and the challenge of identifying fake reviews, a cross-lingual review dataset is created by amalgamating existing publicly available datasets. A fake review recognition model is devised based on the ERNIE2.0 pretrained language model and a semi-supervised generative adversarial network. Initially, ERNIE is employed to extract high-quality linguistic representations of the review data. Next, a generator in a semi-supervised generative adversarial network is utilized to generate noisy data that has a similar distribution to that of the genuine review text data. Finally, the identification of fake reviews is executed in a discriminator. Experimental validation is conducted using the cross-lingual dataset created, and the results indicate that the method achieves a remarkable 81.43% accuracy in identifying fake reviews with only a small amount of labeled data, thereby affirming its effectiveness.