基于PMI-IR的联想词表构造方法研究  

Research on Approach of Thesaurus Construction Based on PMI-IR

作　　者：张泽伟[1] 矫健[1] 张仰森[1] 

机构地区：[1]北京信息科技大学计算机学院智能信息处理研究所,北京100192

出　　处：《计算机技术与发展》2014年第6期140-144,共5页Computer Technology and Development

基　　金：国家自然科学基金资助项目(61070119);北京市属高等学校创新团队建设与教师职业发展计划项目(IDHT20130519);北京市教委专项基金(PXM2012-014224-000020)

摘　　要：通过对大规模查询日志进行挖掘分析进而提高检索的准确率一直是信息检索领域的热点问题。文章提出一种基于PMI-IR(逐点互信息方法)的联想词表构造方法。该方法利用序列模式挖掘算法扫描大规模用户查询日志,获取共现频次超过某一阈值的词组合,进行聚类获取候选同义词集,然后依次计算词wordA与每个候选词的相似度,选择相似度高于某一阈值的词作为词wordA的联想词集,最后形成联想词表。实验表明,借助该方法得到的联想词表进行扩展查询提高了检索的准确率。It has been a hot issue in the field of information retrieval to improve the accuracy of retrieval by mining and analyzing large-scale query logs.A kind of method based on PMI-IR for thesaurus construction is put forward in this paper.The method uses prefix span algorithm to scan the user query log,obtaining these words of which the co-occurrence frequency exceeds a certain threshold,constructing synonym candidate set by cluster.And the similarity of wordA and each candidate word is calculated in turn.These words which are above a certain threshold are selected to construct the synonymy thesaurus.Experimental results show that accuracy can be improved using the thesaurus obtained by the method to extend the search.

关 键 词：逐点互信息方法 联想词表 查询日志 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]