Research on a Federated Learning-Based Attention ERNIE Model for Hierarchical Classification of Short Texts in Power Marketing

Power data is usually recorded in short text form, and traditional management methods are labor-intensive and error-prone, and data sensitivity and privacy issues are also important. In order to extract short text data features effectively, this paper designs a dynamic multi-head sparse attention mechanism based on ERNIE to improve classification performance. At the same time, this paper proposes a multi-center natural language classification training method based on federated learning framework, avoids data sharing through distributed collaboration, and ensures data security by combining differential privacy technology and negative list mechanism. The experimental results show that the method significantly improves the efficiency and accuracy of data processing, while balancing privacy protection and model performance, providing an innovative solution for the intelligent and secure management of power data.