融合汉字部首及多头注意力的中文命名实体识别  

Chinese named entity recognition based on fusion of Chinese radical and multi-head attention

作　　者：薛帆 XUE Fan(School of Information Science and Engineering,Huzhou University,Huzhou,Zhejiang 313000,China)

机构地区：[1]湖州师范学院信息科学与工程学院,浙江湖州313000

出　　处：《计算机应用文摘》2024年第20期136-140,共5页

摘　　要：针对中文命名实体识别准确率不高的问题,提出了一种融合汉字部首与多头注意力机制的中文命名实体识别方法。首先,采用ERNIE预训练语言模型获取词向量表示,并结合汉字部首特征进行融合。将拼接后的向量通过BiLSTM网络提取文本序列的上下文语义信息,同时利用多头注意力机制来增强模型对长序列数据及关键特征的处理能力。最后,通过CRF模型计算输出序列的条件概率分布并进行命名实体识别。实验结果表明,该方法在中文命名实体识别的CLUNER数据集上的表现优于当前主流模型。A Chinese named entity recognition method that integrates Chinese character radicals and multi head attention mechanism is proposed to address the issue of low accuracy in Chinese named entity recognition.Firstly,an ERNIE pre trained language model is used to obtain word vector representations,which are then fused with Chinese radical features.Extract the contextual semantic information of the text sequence from the concatenated vectors through a BiLSTM network,while utilizing a multi head attention mechanism to enhance the model̓s ability to process long sequence data and key features.Finally,the conditional probability distribution of the output sequence is calculated using the CRF model for named entity recognition.The experimental results show that this method performs better than the current mainstream models on the CLUNER dataset for Chinese named entity recognition.

关 键 词：中文命名实体识别 ERNIE模型 部首特征 长短期记忆网络 多头注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]