基于权重标准化SimRank方法的查询扩展技术研究

发布时间：2018-08-05 11:29

【摘要】：查询扩展是信息检索中的一项重要技术。传统的局部分析查询扩展方法利用伪相关文档作为候选词集合,然而部分伪相关文档并不具有很高的相关性。该文利用真实的搜索引擎查询日志,建立了查询点击图,经过多次图结构的转化得到能够反映词之间关联程度的词项关系图,并在图结构的相似度算法SimRank的基础上,提出了一种基于权重标准化的改进SimRank方法,该方法利用词项关系图中词项的全局和间接关系,能够有效挖掘与原始查询相关联的扩展词。同时,为降低SimRank算法的计算复杂度,该文采用了剪枝等策略进行优化,使得计算效率有大幅提高。在TREC标准数据集上的实验表明,该文的方法可以有效地选择相关扩展词。MAP指标较局部分析查询扩展方法提高了1.81%,在P@10和P@20指标评价中效果分别提高了5.44%和3.73%。
[Abstract]:Query extension is an important technology in information retrieval. The traditional local analysis query expansion method uses pseudo-correlation documents as candidate word sets, but partial pseudo-correlation documents are not highly correlated. In this paper, the query click graph is established by using the real search engine query log, and the word item relation graph which can reflect the degree of correlation between words is obtained through the transformation of the structure of multiple graphs, and based on the similarity algorithm SimRank of the graph structure. An improved SimRank method based on weight standardization is proposed. By using the global and indirect relationships of the word items in the word term relational graph, the extended words associated with the original query can be effectively mined. At the same time, in order to reduce the computational complexity of the SimRank algorithm, the pruning strategy is used to optimize the algorithm, which greatly improves the computational efficiency. Experiments on the TREC standard data set show that the proposed method can effectively select the relative extension word .MAP index, which is 1.81% higher than that of the local analysis query extension method, and the effect is 5.44% and 3.73% higher in the evaluation of P10 and Papp20 indexes, respectively.
【作者单位】：大连理工大学信息检索研究室;
【基金】：国家自然科学基金资助项目(60673039,60973068) 国家社科基金资助项目(08BTQ025) 国家863高科技计划资助项目(2006AA01Z151) 教育部留学回国人员科研启动基金高等学校博士学科点专项科研基金资助项目(20090041110002)
【分类号】：TP391.3

【相似文献】