一种结合词汇信息特征的中文命名实体识别方法

Chinese Named Entity Recognition Method Based on Lexical Information Features

在中文命名实体识别任务中,基于字符级嵌入的模型通常仅采用BiLSTM网络获取序列中字符的上下文特征进行实体识别,并没有考虑到词汇信息特征能够在识别实体边界时提供更优的约束.针对此问题,该文提出了一种结合词汇信息特征的中文命名实体识别方法.首先,采用带有残差连接的门控空洞卷积网络提取序列局部特征来表示词汇信息特征,以及采用BiGRU提取序列全局上下文信息特征,并添加句子级注意力机制来增强网络的长序列建模能力;其次,利用稀疏注意力机制对特征进行动态融合,获得包含词汇信息的文本特征;最后,运用CRF学习序列中的约束条件,得到最佳的实体标注结果.对比实验结果表明,该文方法在Resume和CLUENER2020数据集上优于主流的中文命名实体识别方法.

In the task of Chinese named entity recognition,the model based on character-level embedding usually only uses the context features of the characters in the sequence obtained by the BiLSTM network for entity recognition,and does not consider that the lexical information features in the sequence can provide better constraints when identifying the entity boundary.In order to solve this problem,this paper proposes a Chinese named entity recognition method combined with lexical information features.Firstly,the local features of the input sequence are extracted by the gated-dilated convolutional neural network with residual structure to represent the lexical information features,And using BiGRU to extract the context features of the input sequences and add sentence-level attention mechanisms to enhance the long-term sequence modeling ability of the network;Secondly,the sparse attention mechanism is used to dynamically fuse the obtained features to obtain text features containing lexical information;Finally,the constraint conditions in the CRF learning sequence are used to obtain the best entity labeling results.The comparative experimental results show that this method is superior to the mainstream Chinese named entity recognition methods in Resume and CLUSTER 2020 dataset.