网络百科属性抽取.pdfVIP

网络百科属性抽取 + 贾真 , 尹红风, 李天瑞 (西南交通大学 信息科学与技术学院, 成都 610031) Online Encyclopedia Attribute Extraction + JIA Zhen , YIN Hong-Feng, LI Tian-Rui (School of Information and Science Technology, Southwest Jiaotong University, Chengdu 610031, China) + Corresponding author: Phn: +86-28 E-mail:zjia@ , /ccit/ Abstract : The method of extracting class attributes based on Chinese online encyclopedia is proposed in this paper. Unstructured texts of encyclopedia articles are taken as the extracting source. Attribute values are viewed as named entities and the frequent pattern mining technique is applied in the proposed method. The attribute words are generated through mining association rules between word sequences and named entities in frequent patterns. Duplicate attribute words are filtered and uniform attribute names are obtained by the external semantic resource and the computation of the similarity of attribute words. Several classe s of Chinese online encyclopedia entries are downloaded as experimental data. The experimental results indicate that the proposed method is feasible and effective. Key words: attribute extraction; information extraction; online encyclopedia; association rule mining 摘 要: 本文提出了一种基于中文网络百科的属性抽取方法.该方法以中文网络百科自由文本为数据源,将属性 值看作命名实体,对已分类的百科文本进行频繁模式的挖掘.通过分析频繁模式中词语序列与命名实体的关联关系, 得到属性模式与属性词语,借助外部语义资源以及词语相似度计算去除重复的属性词语,得到统一的属性名称.本文 对多个类别的中文百科条目进行了属性抽取实验,实验结果表明了该方法的可行性和有效性. 关键词: 属性抽取;信息抽取; 网络百科;关联规则挖掘 属性是不同类别事物(实体) 的特征.属性知识抽取的两个关键问题是:属性抽取和属性值抽取.网络百科(例 如维基百科,百度百科和互动百科等)是通过网络用户协作方式创建的大百科全书.网络百科涉及的知识类别众 多,各个类别的百科条目中蕴含着丰富的属性知识.基于网络百科进行属性知识抽取对于大规模知识工程的构  Supported by the National Natural Science Foundation of China Director Fund No( 国家自然科学基金委员会主任基 金); Supported by Open Project of Lab. of Management and Control for Complex Systems , Chinese Academy of Sciences No. ( 中国科学院自动化研究所复杂系统管理与