Instruction

This is a simplified version of the core code snippets. After the paper is published, we will publish a complete and runnable code repository.

"transformer_model_fn" is the main function of the model, which calls the training function "get_loss". Compared with the normal transformer based text classification model, our innovation is mainly focused on the "output_layer" function.
In the "output_layer" function, we define the variable "use_user_preference_type". When the variable value is "revision", the Revision-Based Model is implemented, and when the variable value is "representation", the Representation-Based Model is implemented.