CAJ | 학술논문

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。
통과대대규모사순일지진행알굴분석진이제고검색적준학솔일직시신식검색영역적열점문제。문장제출일충기우PMI-IR(축점호신식방법)적련상사표구조방법。해방법이용서렬모식알굴산법소묘대규모용호사순일지，획취공현빈차초과모일역치적사조합，진행취류획취후선동의사집，연후의차계산사wordA여매개후선사적상사도，선택상사도고우모일역치적사작위사wordA적련상사집，최후형성련상사표。실험표명，차조해방법득도적련상사표진행확전사순제고료검색적준학솔。
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 a-bove 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.