吉林大学学报(理学版)

• 计算机 • 上一篇    下一篇

Distant Supervision方法中对齐数据的聚类去噪

朱兆龙, 欧阳丹彤, 叶育鑫   

  1. 吉林大学 计算机科学与技术学院, 符号计算与知识工程教育部重点实验室, 长春 130012
  • 收稿日期:2013-01-31 出版日期:2014-03-26 发布日期:2014-03-20
  • 通讯作者: 欧阳丹彤 E-mail:ouyd@jlu.edu.cn

Clustering Based Denoising for Aligned Data of Distant Supervision

ZHU Zhaolong, OUYANG Dantong, YE Yuxin   

  1. Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education,College of Computer Science and Technology, Jilin Unive
    rsity, Changchun 130012, China
  • Received:2013-01-31 Online:2014-03-26 Published:2014-03-20
  • Contact: OUYANG Dantong E-mail:ouyd@jlu.edu.cn

摘要:

基于聚类和寻找表达语义关系的句子模式, 提出一种基于聚类去噪的distant supervision方法, 解决了传统distant supervision方法在知识库和文本集对齐阶段会引入噪声数据的问题, 得到了一种改进的distant supervision方法. 实验结果表明, 该方法可有效提高关系抽取任务的准确率.

关键词: 关系抽取, 聚类, distant supervision方法, 去噪

Abstract:

A new clustering based denoising method was proposed, with the help of clustering and recognizing patterns of sentences that express a relationship, to reduce the noisy data that was introduced in the alignment step of traditional distant supervision method. The experimental results demonstrate that our approach can significantly improve the accuracy of relation extraction system.

Key words: relation extraction, clustering, distant supervision method, denoising

中图分类号: 

  • TP181