摘要
为解决因外来海洋生物领域实体复杂且实体间存在嵌套导致命名实体识别效果较差等问题,提出基于融合注意力机制的卷积神经网络(CNN)-双向门控循环单元网络(BiGRU)-条件随机场(CRF)网络模型进行外来海洋生物命名实体识别,并构造词向量、词性特征向量等特征作为网络模型的联合输入,以提升网络模型识别效果。结果表明:使用融合多特征向量的CNN-BiGRU-CRF网络模型对外来海洋生物名称实体、时间实体、地名实体3类实体上的命名实体识别结果平均准确率达到了90.62%,平均召回率达到了89.50%,平均F1值达到了90.05%,较传统命名实体识别方法均有较大提高。研究表明,本研究中提出的网络模型可以充分提取文本特征,解决了文本的长距离依赖问题,对外来海洋生物领域的命名实体识别具有较好的识别效果。
In order to solve the problem of poor named entity recognition due to the complex entity and the nesting of entities in the field of exotic marine organisms,the convolutional neural network(CNN)-bidirectional gated recurrent unit network(BiGRU)-conditional random field(CRF)network were used to identify the exotic marine biological entities,and the word vector,part of speech feature vector and other features as the joint input of the network were constructed to improve the recognition effect of the network.Results showed that there was 90.62%of average accuracy of the named entity recognition on the three types of exotic marine biological entities,time entities,and place name entities,the average recall rate of 89.50%,and the average F1 value of 90.05%,which is greatly improved compared to traditional entity recognition methods,using the CNN-BiGRU-CRF network fused with multiple feature vectors.It was found that the network proposed in this study fully extracted and utilized text features,and solved the problem of long-distance dependence of text,with better recognition effect for named entity recognition in the field of exotic marine organisms.
作者
贺琳
张雨
巴韩飞
HE Lin;ZHANG Yu;BA Hanfei(School of Shipping Economics and Management, Dalian Maritime University, Dalian 116026, China)
出处
《大连海洋大学学报》
CAS
CSCD
北大核心
2021年第3期503-509,共7页
Journal of Dalian Ocean University
基金
国家重点研发计划项目(2017YFC1404602)
辽宁省社会科学规划项目(L15BGL040)
大连海事大学基本科研业务费团队项目(3132019353)。
关键词
海洋生物
命名实体识别
卷积神经网络
深度学习
marine organism
named entity recognition
convolutional neural network
deep learning
作者简介
贺琳(1980—),女,副教授。E-mail:helin@dlmu.edu.cn;通讯作者:张雨(1996—),男,硕士研究生。E-mail:Zhangyu270621@163.com。
