SGAN-SAM-ERNIE: A Novel and Effective Detection Scheme for Chinese Fake Reviews

Online reviews play a pivotal role in shaping consumer decisions on e-commerce platforms and social media. Ensuring the authenticity and effectiveness of these reviews is essential for safeguarding consumer rights and preserving platform integrity. Nevertheless, Chinese online reviews pose unique challenges due to their linguistic complexity and the scarcity of annotated data, rendering the efficient identification of fake reviews a persistent challenge.To address these challenges, this study introduces a novel pre-training model called SGAN-SAM-ERNIE, which integrates a self-attention mechanism within a semi-supervised framework. Initially, the ERNIE3.0 pre-training model is utilized to extract high-quality language representations from online review data. Subsequently, the generator within the semi-supervised generative adversarial network (SGAN) generates synthetic data that mimics the distribution of authentic reviews. Meanwhile, the self-attention mechanism is integrated to capture key information and semantic dependencies within the input sequences, thereby enhancing the model's comprehension of the reviews. Finally, a discriminator is employed to distinguish between authentic and fake reviews. Experimental validation was conducted using both cross-lingual and Chinese-specific datasets. The results demonstrate a notable improvement in fake review identification performance, even in scenarios with limited labeled data availability. This underscores the efficacy of the proposed SGAN-SAM-ERNIE model in addressing the challenges associated with identifying fake reviews in Chinese, thereby contributing to a more trustworthy and reliable online review ecosystem.