摘要
为解决现有预训练模型没有充分考虑汉语词汇的分割特征的问题,提出一种基于LEBERT-ILRN-RA-CRF的中文命名实体识别模型。该模型首先利用LEBERT对输入文本进行词汇表增强和嵌入处理,然后利用ILRN模块提取时间信息,并将字符-词融合与字向量采集相结合,最后使用条件随机场(CRF)进行结果校正。此外,该模型还引入了词汇量增强和残差门控注意网络来加强时间特征的提取和局部特征的表达,进一步提升了模型的识别效果。在Weibo和Resume两个数据集上分别取得了71.73%和96.51%的宏F1值。实验表明该模型可以考虑到汉语词汇的分割问题,改善中文命名实体识别任务的识别效果。
In order to solve the problem that the existing pre-trained models do not fully consider the segmentation features of Chinese words,a Chinese named entity recognition model based on LEBERT-ILRN-RA-CRF was proposed.The model firstly used LEBERT to enhance and embed the input text,then used ILRN module to extract time information,combined character-word fusion with word vector acquisition,and finally used conditional random field(CRF)to correct the result.In addition,the model also introduced vocabulary enhancement and residual gated attention network to strengthen the extraction of temporal features and the expression of local features,which further improved the recognition effect of the model.The macro F1 values of 71.73%and 96.51%were obtained on the Weibo and Resume datasets,respectively.Experiments show that the model can take into account the segmentation problem of Chinese vocabulary and improve the recognition effect of Chinese named entity recognition task.
作者
张光明
肖然
张弛
高谦
谈栋华
彭菊红
ZHANG Guangming;XIAO Ran;ZHANG Chi;GAO Qian;TAN Donghua;PENG Juhong(School of Artificial Intelligence,Hubei University,Wuhan 430062,China;School of Computer and Information Engineering,Hubei University,Wuhan 430062,China;Key Laboratory of Intelligent Sensing System and Security(Hubei University),Ministry of Education,Wuhan 430062,China)
出处
《湖北大学学报（自然科学版）》
CAS
2025年第1期118-125,共8页
Journal of Hubei University：Natural Science
基金
湖北省重点研发项目(2023BAB068)资助。
