• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Symposium on Knowledge and Systems Sciences(KSS2004); 20041110-12; Ishikawa(JP) >An Unsupervised Approach to Coreference Resolution
【24h】

An Unsupervised Approach to Coreference Resolution

机译：共监督解析的无监督方法

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Coreference resolution, also known as the process of linking noun phrases (NPs) referring to the same real world entity mentioned in a document, is a difficult and important task in natural language processing. This paper introduces an "incremental" unsupervised Coreference resolution algorithm that can make the most of the transitive property in a Coreference chain as well as the dependencies among noun phrases. These advantages are derived from the observation that the order in which noun phrases are examined is really significant. In our algorithm, at each iteration, we re-rank the order of clustering according to the distinctness based on an entropy estimation. Highly discriminative and confident links between clusters should be processed first to reduce the ambiguity as much as possible and to open up additional useful clues for clustering subsequent hard-to-cluster noun phrases. The experimental evaluation on the MUC-7 corpus demonstrates the advantages over the previous clustering based algorithm and the competitiveness with previous supervised learning methods.
机译：共指解析，也称为链接名词短语（NP）指代文档中提到的同一真实世界实体的过程，在自然语言处理中是一项困难而重要的任务。本文介绍了一种“增量式”无监督共指分解算法，该算法可以充分利用共指链中的传递特性以及名词短语之间的依存关系。这些优势源自观察到名词短语的检查顺序确实很重要的观察结果。在我们的算法中，在每次迭代中，我们都基于熵估计，根据不同性对聚类顺序进行重新排序。应该首先处理集群之间具有高度区分性和信心的链接，以尽可能减少歧义，并为聚类后续难以聚类的名词短语提供更多有用的线索。对MUC-7语料库的实验评估表明，该算法优于以前的基于聚类的算法，并且具有与以前的监督学习方法相比所具有的竞争力。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号