Named Entity Recognition in Fire Control Texts Based on BERT
Xiangru Meng, PengFei Cao, ManNing Ma · 2023
In order to extract named entities from fire control texts more accurately, a named entity corpus based on fire control related texts is constructed. Aiming at the traditional word vector or word vector can not better represent the context semantics and the disadvantages of recurrent neural network in using GPU parallel computing ability, a named entity recognition model of fire control related texts based on BERT is proposed. The word granularity vector matrix is obtained by using the BERT language model, and the use the IDCNN to abstract features from text. The text dependency on a long-distance of sentences is obtained through MA layer. At last, the global best sequence is extracted through CRF. By analyzing the final experimental findings, it is known that the accuracy, recall and$F$value increased 2.78%, 1.59% and 1.33% respectively.