[1]尹 鹏,周 林,郭 强,等.基于短语级注意力机制的关系抽取方法[J].计算机技术与发展,2019,29(09):24-30.[doi:10. 3969 / j. issn. 1673-629X. 2019. 09. 005]
　YIN Peng,ZHOU Lin,GUO Qiang,et al.Relation Extraction Based on Phrase-level Attention[J].,2019,29(09):24-30.[doi:10. 3969 / j. issn. 1673-629X. 2019. 09. 005]
点击复制

基于短语级注意力机制的关系抽取方法()
分享到：

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
29
期数:
2019年09期
页码:
24-30
栏目:
智能、算法、系统工程
出版日期:
2019-09-10

文章信息/Info

Title:
Relation Extraction Based on Phrase-level Attention
文章编号:
1673-629X(2019)09-0024-07
作者:
尹 鹏; 周 林; 郭 强; 刘镇江
江南计算技术研究所,江苏 无锡 214083
Author(s):
YIN Peng; ZHOU Lin; GUO Qiang; LIU Zhen-jiang
Institute of Jiangnan Computing Technology,Wuxi 214083,China
关键词:
关系抽取; 远程监督; 分段卷积神经网络; 注意力机制; TransE 方法; 标签平滑正则化
Keywords:
relation extraction; distant supervision; PCNN; attention mechanism; TransE; label smoothing regularization
分类号:
TP391
DOI:
10. 3969 / j. issn. 1673-629X. 2019. 09. 005
摘要:
关系抽取是自然语言处理的重要研究内容,是知识图谱构建的关键技术。 目前,在神经网络中引入注意力机制进行关系抽取成为主流方法,现有方法一般结合句子单词和实体相关性计算注意力,没有考虑短语和实体关系之间的相关性,并且对实体信息利用不够充分。 针对该问题,提出基于短语级注意力机制的关系抽取方法。 首先用卷积层对词向量做卷积,以滑动窗口的方式得到短语级的向量表示,然后利用短语与实体关系之间的相关性计算注意力。 为了使实体信息利用更充分,用卷积层和池化层分别提取实体短语的深度特征表示,并引入 TransE 的思想表示两个实体关系的特征。最后,采用分段池化方法得到深度特征。 为了减少远程监督中错误标签的干扰,使用标签平滑正则化(LSR)把原来的“硬冶标签改为“软冶标签。 实验结果表明,该方法能够有效利用短语信息和实体关系信息,对实体关系抽取效果有较大的提升。
Abstract:
Relation extraction (RE) is an important research content of natural language processing and a key technology of knowledge graph construction. Relation extraction using neural network with attention has become the mainstream method. However,existing methods usually combine words and entities to calculate attention,not take into account the correlation between phrase and entity relations,and not make enough use of entity information. For this,phrase-level attention model is proposed. Convolution layer is used to generate phrase level feature from embedded word vector,and correlation between phrase and entity relationship applied to calculate attention. In order to make full use of entity information,convolution layer and max pool layer is used to extract depth feature of entity phrase,as well as TransE method introduced to represent the characteristics of two entity relations. Finally,piecewise max pooling is used to get depth features. In order to reduce interference of wrong label in distant supervision,label smoothing regularization (LSR) is used to change the original “hard”labels to “soft” labels. Experiment shows that this model can ffectively utilize phrase and entity relationship information,and significantly improves the performance of relation extraction.

相似文献/References:

[1]田 芳,孙 晓,孙本旺.基于网络知识百科的情感语义抽取研究[J].计算机技术与发展,2019,29(06):52.[doi:10. 3969 / j. issn. 1673-629X. 2019. 06. 011]
　TIAN Fang,SUN Xiao,SUN Ben-wang.Research on Extraction of Affective Lexical Semantic Using Encyclopedia of Network Knowledge[J].,2019,29(09):52.[doi:10. 3969 / j. issn. 1673-629X. 2019. 06. 011]
[2]崔从敏,施运梅,袁 博,等.面向政府公文的关系抽取方法研究[J].计算机技术与发展,2021,31(12):26.[doi:10. 3969 / j. issn. 1673-629X. 2021. 12. 005]
　CUI Cong-min,SHI Yun-mei,YUAN Bo,et al.Research on Relation Extraction Method for Government Documents[J].,2021,31(09):26.[doi:10. 3969 / j. issn. 1673-629X. 2021. 12. 005]
[3]盖素丽,檀改芳,唐 松,等.面向科技服务的专家推荐模型研究[J].计算机技术与发展,2022,32(10):21.[doi:10. 3969 / j. issn. 1673-629X. 2022. 10. 004]
　GE Su-li,TAN Gai-fang,TANG Song,et al.Expert Recommendation Model for Technology Services[J].,2022,32(09):21.[doi:10. 3969 / j. issn. 1673-629X. 2022. 10. 004]
[4]张 鑫,冼广铭*,梅灏洋,等.基于 Span 方法和多叉解码树的实体关系抽取[J].计算机技术与发展,2023,33(05):152.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 023]
　ZHANG Xin,XIAN Guang-ming*,MEI Hao-yang,et al.Entity Relation Extraction Based on Span Method and Multi-fork Decoding Tree[J].,2023,33(09):152.[doi:10. 3969 / j. issn. 1673-629X. 2023. 05. 023]
[5]卜意磊,庞文迪,吴甜甜,等.面向食品监管领域的知识图谱构建研究[J].计算机技术与发展,2023,33(06):202.[doi:10. 3969 / j. issn. 1673-629X. 2023. 06. 030]
　BU Yi-lei,PANG Wen-di,WU Tian-tian,et al.Research on Knowledge Graph Construction for Food Supervision[J].,2023,33(09):202.[doi:10. 3969 / j. issn. 1673-629X. 2023. 06. 030]
[6]付学敬,丁肖摇.基于多层次语义感知的中文关系抽取研究[J].计算机技术与发展,2024,34(01):143.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 021]
　FU Xue-jing,DING Xiao-yao.Research on Chinese Relation Extraction Based on Multi-level Semantic Perception[J].,2024,34(09):143.[doi:10. 3969 / j. issn. 1673-629X. 2024. 01. 021]
[7]姚飞杨,刘晓静*.基于 RoBERTa-Effg-Adv 的实体关系联合抽取方法[J].计算机技术与发展,2024,34(03):147.[doi:10. 3969 / j. issn. 1673-629X. 2024. 03. 022]
　YAO Fei-yang,LIU Xiao-jing*.Entity and Relation Joint Extraction Method Based on RoBERTa-Effg-Adv[J].,2024,34(09):147.[doi:10. 3969 / j. issn. 1673-629X. 2024. 03. 022]

更新日期/Last Update: 2019-09-10