BERT-based military field composite named entity recognition method

The invention provides a BERT-based military field composite named entity recognition method, which comprises the following steps of: representing a word vector by utilizing a BERT pre-training model in an input layer, representing a word vector by utilizing Word2Vec, performing Word Embedding by combining the word vector and the word vector, then introducing data enhancement operation, splicing on a word vector representation layer to enhance original input information, and constructing a sentence initial vector; using Bi-On-LSTM (Long Short Term Memory) to capture global semantic information of the text in a coding layer; introducing an attention layer, and updating a semantic weight; an LSTM Unit long-short-term memory network is adopted in a decoding layer, nested named entity extraction is carried out, the number of output labels is not limited to be single any more, the advantages of traditional On-LSTM in the aspect of sentence level information extraction are inherited, Softmax is used for predicting