Expanding the Horizons of Chinese Text Classification with Federated Learning Across Domains

Federated learning (FL) is a new technology that can correct privacy issues in machine learning by training models on dispersed units. Through applying a hybrid algorithm comprising Federated Learning (FedLearning), Local Batch Normalization (LBN), and Parameter Factorization Weighting (PFW), the article now introduces a Chinese text classifier called the FedBN-PW-CTC Model. This model intends to ensure that accuracy and F1-value substantially improve when solving text categorization problems. Taking into account the local batch normalization, the weighting of the models, and the federated learning with group leaders is the next in line; this approach assumes the priority of a closer consideration of the different elements in Chinese news content and keeping the information secret. We determine the equivalence of the structural characteristics of Siamese networks and PPG signal databases without and with various patients. The model also offers robust grades of precision and accuracy, accompanying its successful scores. The model is superior to the literature in automating text production with a $94 \%$ F1 score, as indicated by the statistical data and the site accuracy of $95 \%$. For Bank A and B, the confidence level is $94 \%$ and $95 \%$, respectively, and the margin of error is $6 \%$ and $5 \%$.Moreover, the p-values for the accuracy and F1 score comparisons were also calculated to be <0.05. Results indicate the highest increase is most likely. Following the test results, the suggested model can be used to solve other similar privacy-sensitive data processing tasks, such as learning Chinese text classification.