<p>E-commerce platforms face the issue of spammer groups that post many fraudulent reviews within a certain period. This practice misleads consumers and undermines fair competition among merchants. Researchers have proposed various methods to combat these spammers. However, these methods typically handle user feature representation and candidate group division independently, lacking an effective feedback mechanism between the two. Moreover, the vast discrepancy in the number of negative samples compared to positive samples in the data leads to reduced recognition accuracy in detection models, affecting the precision of detection outcomes. Therefore, we propose a spammer group detection method based on self-supervised deep clustering. Initially, the relationship between user review timing and product ratings is extracted from user review data, and user relevance is calculated and used as weights to construct a weighted user relationship graph. Subsequently, we integrate deep learning models with clustering algorithms to propose a self-supervised deep clustering model that jointly optimizes user representation and clustering distribution. This model employs graph and node autoencoders to capture global structural information and local preference information of user nodes, respectively, and designs a linear fusion method to enhance user feature representation. Additionally, we construct a reliable target distribution and introduce Kullback-Leibler(KL) divergence to form a self-supervised mechanism, continuously optimizing feature representation and clustering assignment to refine high-quality candidate groups. Finally, we propose an anomaly detection method based on the Gaussian Mixture Model (GMM), which designs a filtering mechanism to improve the detection efficiency of spammer groups. Experiments indicate that the proposed method outperforms baseline methods on the Amazon, Yelp, and YelpChi datasets.</p>