Method for carrying out text classification on federated learning large model based on shallow feature pre-training

The invention discloses a method for carrying out text classification on a federated learning large model based on shallow feature pre-training. The method comprises the steps that a server side obtains a global model; the client obtains an initial parameter of the global model, constructs a local model according to the initial parameter of the global model and a to-be-trained processing layer number l specified by the server, and trains the local model; the client uploads the updated parameters of the lth processing layer and the parameters of the output layer to the server for aggregation to obtain updated parameters, updates a global model of the server, and issues the updated parameters to each client for a new round of federal learning training; the client obtains the updated parameters and a to-be-trained processing layer number l'newly specified by the server from the server, and federal learning is repeated; and after federal learning is completed, the server replaces the corresponding parameters of t