Data Quality-Aware Federated Learning for Fake Review Detection

Fake reviews are a significant challenge for online consumer and social media platforms, as they mislead consumers and disrupt fair market competition. Traditional centralized detection methods face numerous limitations in handling massive data volumes, heterogeneous distributions, and privacy compliance. To address these issues, this paper proposes a data quality-aware federated learning framework for efficiently detecting fake reviews while preserving the privacy of data holders. Specifically, we introduce a data quality evaluation module during the federated learning training process. This module quantifies the quality of each client's data based on multiple metrics, including annotation accuracy, textual completeness, and user behavior. Higher-quality data clients are assigned greater weights during global model aggregation, thereby mitigating the interference of low-quality or noisy data. Experimental results demonstrate that compared to traditional federated averaging and other baseline methods, our approach significantly improves accuracy, precision, and recall. Additionally, it exhibits greater robustness and faster convergence under various non-independent and identically distributed (non-IID) and multi-noise scenarios. Furthermore, differential privacy ensures the security of participants' raw data and quality evaluation information, enhancing the feasibility of real-world deployment.