Robust teaching knowledge point identification method and device based on adversarial learning

The invention relates to a robust teaching knowledge point identification method and device based on adversarial learning. The method is improved in three aspects: 1) data production: processing, summarizing and concluding electronic textbook data of a computer network to construct a knowledge point identification data set of computer network subjects, and taking the knowledge point identification data set as input of a model; 2) model improvement: carrying out model structure modification on the basis of an ALBERT and TextCNN combined model, combining information of an entity in an input sentence in hidden layer output of the ALBERT, and improving the capability of capturing the entity information of the model; and 3) method improvement: carrying out new improvement on the training process of the model, adding adversarial training, and effectively improving the generalization ability and robustness of the model. According to the method, a combination model of ALBERT and TextCNN is applied to a knowledge poin