摘要
针对上下位关系在分类层级结构建立阶段遇到的多义性问题,给出一种概念空间中上下位关系意义识别的方法.单个概念的意义识别问题被转换为概念空间中上下位关系的意义识别.首先利用并列语境解决语境稀疏问题,获取上下位关系意义的语境.然后利用《同义词词林》对每个语境进行词义修正,以三种特征计算特征词权重,构建"关系-词"的高维向量空间,然后通过潜在语义分析降维,获取上下位关系意义的潜在语义,最后组平均聚类后得到关系的意义划分.在实验中,给出了聚类阈值自动调整函数,分析了词林和潜在语义分析的作用,实验结果证实了方法的有效性.
For the polysemy of hyponymy in the phase of building taxonomic hierarchy, this paper presents a method of sense recognition of hyponymy based on concept space. The problem of sense recognition of single concept is transformed into recognition of hyponymy in concept space. Firstly, the contexts of hyponymy are acquired iteratively using coordinate relation patterns. Secondly CiLin and the weight of feature words are used to construct a hyponymy-word vector space. Then LSA is used to reduce the dimension of the vector space. In the final phase, the senses of hyponymy can be recognized using average-group clustering. The relation of decreasing degree of similarity and threshold of clustering, and the effect of CiLin and LSA in experiment are analyzed. Experimental results show that the method is adequate of partitioning the senses the hyponymy.
出处
《计算机学报》
EI
CSCD
北大核心
2009年第8期1651-1661,共11页
Chinese Journal of Computers
基金
国家自然科学基金(60573064
60705022
60773059)
国家"八六三"高技术研究发展计划项目基金(2007AA01Z325)
北京工业大学博士启动基金(X0006014200803)资助
关键词
知识获取
上下位关系
潜在语义分析
关系获取
概念空间
意义聚类
knowledge acquisition
hyponymy relation
latent semantic analysis
relation acquisition
concept space
sense clustering
