Chinese Fine-grained Named Entity Recognition based on Deep Learning

For applications such as intelligent retrieval systems and intelligent question answering systems, named entity recognition is a fundamental and crucial sub-task.The fine-grained named entity recognition task, as opposed to the conventional coarse-grained named entity recognition task, is more in line with the requirements of real-world applications and requires models with better feature extraction capabilities. Most of the current Chinese fine-grained named entity recognition models only consider the semantic aspects of the text and ignore its spatial features. To address this issue, this paper proposes a deep learning-based Chinese fine-grained named entity recognition model that uses the BERT pre-trained language model as the encoding layer to vectorize the input sequence to encode, then combines BiGRU and IDCNN into a feature extraction layer to extract features from different angles, and finally sends the results to the CRF decoding layer to compute the optimal sequence and decode it using the Viterbi algorithm. In addition, the ELU function replaces the ReLU activation function of the conventional IDCNN network to improve the overall training effect of the model. Experiments show that the model performs well on the Chinese fine-grained named entity recognition dataset CLUENER2020, as well as on three public datasets: People's Daily dataset, MSRA dataset and Electronic Resume dataset. The training procedure of this method is stable, and it effectively recognises Chinese fine-grained named entities and exhibits excellent generality.