全部 标题 作者
关键词 摘要

OALib Journal期刊
ISSN: 2333-9721
费用:99美元

查看量下载量

相关文章

更多...

一种双向长短时记忆循环神经网络的问句语义关系识别方法
Identifying semantic relations in users’ questions by using bi-directional long short term memory neural networks

DOI: 10.7631/issn.1000-2243.16432

Keywords: 语义关系识别 长短时记忆 循环神经网络 梯度弥散 语义特征
semantic relation identification long short term memory recurrent neural network gradient diffusion semantic feature

Full-Text   Cite this paper   Add to My Lib

Abstract:

提出一种基于双向长短时记忆循环神经网络的问句语义关系识别方法. 利用循环神经网络直接从词学习问句的语义特征表示,不需要自然语言处理工具进行特征抽取,有效避免了误差传递问题. 同时,在网络中加入双向结构和长短时记忆模块,有效改善传统循环神经网络在训练过程中的“梯度弥散”问题. 加入基于主实体位置的分段最大池化操作,相对于传统单一最大池化,能保留问句文本中的有效语义特征. 通过在电力领域真实问题集上实验比较,本方法相对于传统方法能有效提升问句语义关系识别的性能,问句语义关系分类结果F1值提高4.5%.
This paper propose a novel approach based on bi-directional long short term memory neural networks. Compared with traditional methods,our approach could directly learn the feature representation from the words directly and do not need any NLP operation. In this way,the problem of error propagation could be avoided. And we apply bi-directed structure and long short term memory module in our network. Moreover,we design a main entity aware segment-based max pooling operation in our network. As a result,the features for semantic relation in the questions could be captured more precisely. The experimental results on real data in the electronic domain show that our approach could outperform state-of-the-arts,and the F1 value could be improved by 4.5%

Full-Text

Contact Us

service@oalib.com

QQ:3279437679

WhatsApp +8615387084133