Identification of Malicious E-commerce Users Based on User Rating Behavior and GNN

With the rapid rise of e-commerce, authentic user evaluations are particularly important in purchasing decisions. The increase in fake evaluations has made effective identification a top priority. The study addresses this challenge by introducing a reputation strategy that combines scoring patterns and differences, GNR metrics, and adversarial data augmentation technology, to improve the effectiveness of fraud detection. The study conducted experiments using three datasets, Netflix, Movielens2, and Movielens_100, which record user ratings of movies at different scales. The main performance metrics include recall, F1 value and area under the curve (AUC). The experiment showed that when the proportion of fraudulent users was 0.035, the recall value of reputation ranking technology strategy based on user evaluation mode and bias was 0.79, with stability exceeding 0.990. After applying the GNR metrics, the deviation ranking method showed a significant reduction in peak user count and improved the overall performance by 9.40%. The accuracy of the iterative group-based ranking and the iterative balance ranking increased by 2.89% and 2.54%, respectively. After introducing the adversarial data augmentation technology, the fraud detector based on graphical neural networks improved recall and F1 by 1.20% and 1.34%, respectively compared with the disguised fraudster model in the case of data scarcity. It can be seen that the method combining multiple strategies and technologies demonstrates improved performance in e-commerce user evaluation fraud detection, far surpassing traditional methods. This study has brought significant significance and value to the e-commerce field.