Research on Chinese Named Entity Recognition Based on RoBERTa-BIGRU-MRC Model
Huai Peng, Xianghong Tang · Atlantis Highlights in Intelligent Systems/Atlantis highlights in intelligent systems · 2022
Nested entity is the focus and difficulty of Chinese named entity recognition.The existing methods regard nested NER as two subtasks of Chinese word segmentation and sequence annotation.This method depends very much on the quality of input word vector, and low-quality word vector will lead to the error propagation of the model.To solve the above problems, a Chinese named entity recognition model based on RoBERTa-BiGRU-MRC is proposed.Firstly, RoBERTa is used to embed the entity type description and sentences to obtain the dynamic word vector.Secondly, BiGRU is used to extract contextual semantic features for further understanding of semantic information.Then two binary classifiers are constructed to better predict the probability value of the index at the beginning and end of the entity.Finally, the accuracy of the model is improved by constructing the loss function optimizer of predicted value and real value.Experiments were conducted on Chinese MSRA and onto4 data sets, and the accuracy, recall and F1 value were used as evaluation indexes.The experimental results show that the F1 value of the optimized model is 0.41 and 0.36 higher than that of the traditional sequence annotation model, respectively.