Chinese Named Entity Recognition Based on Character Level Multi Feature Fusion

For the task of named entity recognition, this paper proposes a character-level multi-feature fusion named entity recognition method based on the IDCNN model, which can adjust the receptive field range through parameters. A distributed representation method combining radicals and character features is proposed to solve the problem of insufficient representation of Chinese characters' latent features. The character-level feature representation is input into IDCNN for feature extraction. IDCNN can make full use of the parallel ability of GPU under the premise of extracting long-distance semantic information, and finally improve the accuracy of entity label prediction through the CRF layer. In order to verify the effectiveness of the model, this paper conducts experiments on the commonly used MSRA data sets and Resume data sets. The experimental results show that the results of the model on MSRA and Resume data sets surpass those of Lattice LSTM, LR-CNN, and PLET. Compared with the PLET model with the best results, the F1 value of the model in this paper is increased by 0.55% and 0.03%, respectively.