Literature new word discovery method and system based on mask language model

The invention discloses a literature new word discovery method and system based on a mask language model, and belongs to the technical field of artificial intelligence natural language processing, the method employs a mask language training component, a model dependency relationship operation component and a maximum probability operation component, the mask language training component carries out data cleaning and sentence segment segmentation on literature data; an Attention mechanism and a feedforward neural network are constructed through a training composition vector identifier Word Embedding of the training set, combining the Attention mechanism and the feedforward neural network into a group of Encoder, and constructing an Encoder training model; the coded training set is subjected to random shielding, part of the input token is used as training set input, the shielded token is used as output, the mode is used as a data generator, and the deep bidirectional representation network is trained. According t