融合局部共现和上下文相似度的查询扩展方法

The method of query expansion based on local co-occurrence and context similarity

为解决信息检索中用户查询可能与索引文档信息表示不匹配从而影响检索效果的问题,提出一种融合局部共现和上下文相似度的查询扩展方法,从与查询词具有共现关系的邻接词和与查询词具有高相关性或同指关系的词两个方面对用户输入查询词进行扩展,重点测试邻接词的取词窗口大小以及上下文向量的最优长度。试验表明:与采用单一扩展方法相比,融合方法的平均准确率取得了明显提高,当邻接词的窗口大小取5,上下文向量的长度取15时,具有更好的平均准确率。

In order to solve the mismatch between user query and representation of index document, which affects the performance of information retrieval, this paper proposes a query expansion method based on the integration of local co- occurrence and context similarity. The method expands the user query through the following two ways, one is the adja- cent words which have co-occurrence relationship with the query words, the other is the similar words which have a high correlation with the query words. The method focuses on testing the influence of the adjacent words＇ window size and the optimal length of context vectors. Experimental results show that compared with the single expansion method, our method can improve the average accuracy obviously, and the average accuracy reaches the highest when the window size of adjacent words is 5, and the length of the context vector is 15.