基于加权SimRank的中文查询推荐研究

发布时间：2018-12-12 13:21

【摘要】：查询推荐是搜索引擎系统中的一项重要技术,其通过推荐更合适的查询以提高用户的搜索体验。现有方法能够找到直接通过某种属性关联的相似查询,却忽略了具有间接关联的语义相关查询。该文将用户查询及查询间直接联系建模为查询关系图,并在图结构相似度算法SimRank的基础上提出了加权SimRank(简称WSimRank)用于查询推荐。WSimRank综合考虑了查询关系图的全局信息,因而能挖掘出查询间的间接关联和语义关系。然而,WSimRank复杂度太高而难以实用,该文将WSimRank转换为一个状态层次图的遍历和计算过程,进而采用动态规划、剪枝等策略对其进行优化从而可以实际应用。在大规模真实Web搜索日志上的实验表明,WSimRank在各项评价指标上均优于SimRank和传统查询推荐方法,其MAP指标接近0.9。
[Abstract]:Query recommendation is an important technology in search engine system. The existing methods can find similar queries directly associated with some attributes, but ignore semantic queries with indirect association. In this paper, the direct relation between user query and query is modeled as query relational graph, and based on the similarity algorithm of graph structure SimRank, a weighted SimRank (WSimRank) is proposed to query recommendation. WSimRank considers the global information of query diagram. Therefore, the indirect association and semantic relation between queries can be mined. However, the complexity of WSimRank is too high to be practical. In this paper, the WSimRank is transformed into the traversal and computation process of a state hierarchy graph, and the dynamic programming and pruning strategies are adopted to optimize it so that it can be applied in practice. Experiments on large scale real Web search log show that WSimRank is superior to SimRank and traditional query recommendation method in evaluating index, and its MAP index is close to 0.9.
【作者单位】：中国科学院计算技术研究所;中国科学院研究生院;
【基金】：国家自然科学基金资助项目(60603094) 北京市自然科学基金资助项目(4082030) 国家973资助项目(2007CB311103) 国家863计划资助项目(2006AA010105)
【分类号】：TP391.3

【相似文献】