Wikipedia Relatedness Measurement Methods and Influential Features
Wikipedia Relatedness Measurement Methods and Influential Features
复制标题
DOI：
10.1109/waina.2009.206
复制
发表时间：
2009-05
期刊：
影响因子：
--
通讯作者：
Kotaro Nakayama;Masahiro Ito;T. Hara;S. Nishio
中科院分区：
文献类型：
--
作者：
Kotaro Nakayama;Masahiro Ito;T. Hara;S. Nishio
作为知识提取的语料库，维基百科因其概念覆盖范围广、准确性高、分析结构易于处理等特点，已成为NLP、WWW、IR、AI等各个领域研究人员的热门资源之一。概念间的关联度量是维基百科分析的传统研究课题之一。相关性测量研究的价值已被广泛认可，因为它在IR中的查询扩展和WSD（词义消歧）中的上下文识别等方面有着广泛的应用。已经提出了许多方法，它们证明了有许多特征可以用来衡量维基百科中概念之间的相关性。在过去的研究中，使用了分类、词（链接）共现、页间链接、Infoboxes等特征来达到这一目的。然而，对于这些分散的特征，似乎缺乏一个集成的特征选择模型，因为我们仍然不清楚哪个特征是有影响的，以及我们如何将它们集成以达到更高的精度。本文是一篇立场论文，提出了一种基于支持向量回归（SVR）的综合特征选择模型，研究每个特征的影响，寻求一个高精度和高覆盖率的特征组合模型。
As a corpus for knowledge extraction, Wikipedia has become one of the promising resources among researchers in various domains such as NLP, WWW, IR and AI since it has a great coverage of concepts for wide-range domain, remarkable accuracy and easy-handled structure for analysis. Relatedness measurement among concepts is one of the traditional research topics on Wikipedia analysis. The value of relatedness measurement research is widely recognized because of the wide range of applications such as query expansion in IR and context recognition in WSD (Word Sense Disambiguation). A number of approaches have been proposed and they proved that there are many features that can be used to measure relatedness among concepts in Wikipedia. In the past, previous researches, many features such as categories, co-occurrence of terms (links), inter-page links and Infoboxes are used to this aim. What seems lacking, however, is an integrated feature selection model for these dispersed features since it is still unclear that which feature is influential and how can we integrate them in order to achieve higher accuracy. This paper is a position paper that proposes a SVR (Support Vector Regression) based integrated feature selection model to investigate the influence of each feature and seek a combine model of features that achieves high accuracy and coverage.