Chinese Resume Named Entity Recognition Based on ChineseBERT

The existing methods for recognizing named entities in Chinese electronic resumes have been found to have low accuracy. While the BERT pre-training model has been demonstrated to achieve higher accuracy compared to traditional models, the masking language modeling task may result in the random masking of a part of a single or consecutive word in the Chinese language environment. This practice may result in the loss of complete word information. In other words, when masking, the entire word is masked instead of a portion of the characters, which can better preserve the semantic integrity of Chinese. The Chinese BERT pre-training model is employed to embed words in the input text, thereby obtaining dynamic word vectors and resolving the issue of polysemy. BiLSTM is utilized to obtain contextual structural features, which are employed to deeply my semantic relations. The spliced vectors are then input to the CRF layer for Viterbi decoding, which enables the learning of constraint relations between labels and the subsequent output of the correct labels. The experimental results demonstrate that the method achieves an F1 value of 94.86% in the Resume e-resume dataset.