Remote supervision and supervised relation classification method

According to the relation classification method combining remote supervision and supervision, the teacher model is obtained through supervision data training, the entity is randomly covered when the remote supervision data is used for training the student model, meanwhile, the supervision signal of the teacher model is introduced through the distillation technology, and therefore the purpose of noise reduction is achieved. After the student model is obtained, the student model is finely adjusted on the supervised data, and the supervised data is further utilized to enhance the student model. 本发明提供了一种结合远程监督和有监督的关系分类方法，通过有监督数据训练得到教师模型，在使用远程监督数据训练学生模型时对实体进行随机遮掩，同时利用蒸馏技术引入教师模型的监督信号，从而达到降噪的目的。得到学生模型后，在有监督数据上微调学生模型，进一步利用有监督数据增强学生模型。