Privacy Preserving Approach to Communication Data Mining Based on Federated Learning

Current traditional privacy protection methods for communication data achieve functions such as anti-tampering and resistance to single point of failure attacks through edge computing, which leads to poor data protection due to the lack of aggregation processing of user behavior. In this regard, a privacy protection method for communication data mining based on federated learning is proposed. A secure aggregation protocol is constructed to aggregate user interaction behaviors, and a user sensitivity scoring mechanism is constructed to establish a privacy protection model for communication data mining. In the experiments, the privacy-preserving effect of the proposed method is verified. The analysis of the experimental results shows that the data privacy protection model constructed by the proposed method has a high degree of convergence of the model and has a more excellent data privacy protection performance.