基于双流特征互补的嵌套命名实体识别

Identification of nested named entities based on dual-flow features complementation

针对以往句子在文本编码后不能获得高效的特征信息,提出一种基于双流特征互补的嵌套命名实体识别模型。句子在嵌入时以单词的字级别和字符级别两种方式嵌入,分别通过神经网络Bi-LSTM获取句子上下文信息,两个向量进入低层级与高层级的特征互补模块,实体词识别模块和细粒度划分模块对实体词区间进行细粒度划分,获取内部实体。实验结果表明,模型相较于经典模型在特征提取上有较大的提升。

Nested named entity recognition model based on dual stream feature complementation was proposed to solve the problem that the previous sentences can not obtain efficient feature information after text coding.Sentences were embedded in two ways of word level and character level.The sentence context information was obtained through the neural network BiLSTM.The two vectors entered the feature complementation module at the low level and the high level.The entity word recognition module and the fine-grained partition module finely partitioned the entity word interval to obtain the internal entities.Experimental results show that the model has great improvements in feature extraction compared with the classical model.