上海交通大学学报

上海交通大学学报（自然版）

• 自动化技术、计算机技术 • 上一篇    下一篇

基于转移学习的命名实体挖掘技术

翟海军1,2，郭勇3，郭嘉丰2，程学旗2
  

  1. (1. 中国科学技术大学 计算机科学与技术学院， 合肥 230027； 2. 中国科学院计算技术研究所， 北京 100190； 3. 北京系统工程研究所， 北京 100101)
  • 收稿日期:2010-03-02 修回日期:1900-01-01 出版日期:2011-02-28 发布日期:2011-02-28

A Named Entity Mining Method Based on Transfer Learning

ZHAI Haijun1,2,GUO Yong3,GUO Jiafeng2,CHENG Xueqi2   

  1. (1. School of Computer Science and Technology, University of Science and Technology of China, Hefei 230027, China; 2. Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, China;3. Beijing Institute of System Engineering, Beijing 100101, China)
  • Received:2010-03-02 Revised:1900-01-01 Online:2011-02-28 Published:2011-02-28

摘要： 研究了针对大规模查询日志中丰富的命名实体的挖掘技术，通过利用Wikipedia数据，结合转移学习方法构建目标类别的分类器.该技术很好地利用了监督学习的优越性能以提高查询日志中命名实体挖掘的准确性，同时也解决了监督学习方法中大规模标注的问题.实验结果表明，基于转移学习的命名实体挖掘技术具有优越的命名实体挖掘性能.

关键词: 转移学习, 命名实体挖掘, 正例学习

Abstract: This paper addresses the problem of mining named entities from query logs. A novel scheme was introduced based on transfer learning, which trains classifier for target category by leveraging Wikipedia data source. In this way it can greatly make use of supervised learning and also deal with the large scale labeling problem. The experiment results show the effectiveness of the novel scheme based on transfer learning.

中图分类号: