软件导刊2025，Vol.24Issue(6)：1-9,9.DOI:10.11907/rjdk.241378
基于深度学习与特征融合的命名实体识别
Named Entity Recognition Based on Deep Learning and Feature Fusion
摘要
Abstract
In recent years,deep neural network recognition methods based on representation learning,such as character embedding and word embedding,have achieved good recognition results.However,existing methods largely ignore or oversimplify the correlation between features at different levels,resulting in a large amount of information being lost.For named entity recognition,a model based on convolutional neural network(CNN)is proposed.The model calculates the importance of each embedded feature through adaptive weight mechanism,and then merges these features effectively to enhance the representation ability of the model.A multi-layer feature fusion strategy is designed to use the hierarchical structure of CNN to dig the local and global information in text data and the complex relationship between them.In addition,the model also introduces attention mechanism to further optimize the adaptive learning process of feature weights,so that the model can automati-cally identify and focus on more task-critical features.Experimental evaluations on three publicly available datasets(CONLL2003,JNLPBA,and NCBI-Disease)show that the model captures richer semantic features,outperforms a set of state-of-the-art baseline methods,and has an F1 score of up to 94.67%in the NCBI-Disease dataset.关键词
命名实体识别/卷积神经网络/特征融合/自适应权重/注意力机制Key words
named entity recognition/convolutional neural network/feature fusion/adaptive weight/attention mechanism分类
信息技术与安全科学引用本文复制引用
朱琛,徐艺武,冯嘉美..基于深度学习与特征融合的命名实体识别[J].软件导刊,2025,24(6):1-9,9.基金项目
国家自然科学基金项目(61672389) （61672389）
广州市大数据智能教育重点实验室项目(201905010009) （201905010009）
