基于标签迁移和深度学习的跨语言实体抽取研究  

Research on Cross-lingual Entity Extraction Based on Tag Transfer and Deep Learning

作　　者：余传明[1] 黄婷婷 林虹君 安璐[3] Yu Chuanming;Huang Tingting;Lin Hongjun;An Lu(School of Information and Safety Engineering,Zhongnan University of Economics and Law,Wuhan 430073,China;School of Statistics and Mathematics,Zhongnan University of Economics and Law,Wuhan 430073,China;School of Information Management,Wuhan University,Wuhan 430072,China)

机构地区：[1]中南财经政法大学信息与安全工程学院,湖北武汉430073 [2]中南财经政法大学统计与数学学院,湖北武汉430073 [3]武汉大学信息管理学院,湖北武汉430072

出　　处：《现代情报》2020年第12期3-16,35,共15页Journal of Modern Information

基　　金：国家自然科学基金面上项目“面向跨语言观点摘要的领域知识表示与融合模型研究”(项目编号:71974202)。

摘　　要：[目的/意义]从跨语言视角探究如何更好地解决低资源语言的实体抽取问题。[方法/过程]以英语为源语言,西班牙语和荷兰语为目标语言,借助迁移学习和深度学习的思想,提出一种结合自学习和GRU-LSTM-CRF网络的无监督跨语言实体抽取方法。[结果/结论]与有监督的跨语言实体抽取方法相比,本文提出的无监督跨语言实体抽取方法可以取得更好的效果,在西班牙语上,F1值为0.6419,在荷兰语上,F1值为0.6557。利用跨语言知识在源语言和目标语言间建立桥梁,提升低资源语言实体抽取的效果。[Purpose/Significance]This paper explores how to better solve the entity extraction problem of low resource languages from a cross-lingual perspective.[Method/Process]With English as the source language,Spanish and Dutch as the target language,an unsupervised cross-lingual entity extraction method combining self-learning method and GRU-LSTM-CRF network is proposed based on the idea of transfer learning and deep learning.[Result/Conclusion]Compared with the supervised cross-lingual entity extraction method,the unsupervised cross-lingual entity extraction method proposed in this paper can achieve better results.In Spanish,the value of F1 is 0.6419,and in Dutch,the value of F1 is 0.6557.Cross-lingual knowledge is used to build a bridge between source language and target language to improve the effect of entity extraction of low-resource languages.

关 键 词：知识获取 实体抽取 跨语言 深度学习 标签映射 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]