基于线性回归的相关查询推荐  

Related query recommendation based on linear regressio

作　　者：翟海军[1,2] 张刚[2] 张瑾[2] 

机构地区：[1]中国科学技术大学计算机科学与技术系,合肥230027 [2]中国科学院计算技术研究所信息智能与信息安全研究中心,北京100080

出　　处：《高技术通讯》2010年第6期596-601,共6页Chinese High Technology Letters

基　　金：863计划(2006AA010105;2007AA01Z416)资助项目

摘　　要：在分析搜索引擎查询日志的基础上,提出了一种基于线性回归的相关查询推荐方法。该方法考虑了查询串之间的多种关联关系,包括查询串会话共现、点击页面共享和查询串文本相似性,以避免因直接应用查询串之间的部分关联关系造成易受查询日志中噪音的影响。相比以往的方法,采用线性回归的方法来识别相关查询推荐的有效特征,能更好地解决噪音问题和进行有效的相关查询推荐。实验证实,采用线性回归挖掘的识别中文相关查询串的特征非常有效,且所提出的回归模型的预测准确率优于现有的方法。In this paper linear regression is applied to related query recommendation. A novel linear regression model is proposed to recommend related queries from query logs. In this model many types of association relationships, including query session cooccurence, URL-clicked sharing and text similarity, are identified and leveraged, to avoid the circumstance that part of these relationships which are direcdy applied may be largely affected by the noise in query logs. In this work the linear regression is proposed to identify effective features, and this method can effectively deal with the noise problem. The experiments demenstrate that the features identified with linear regression are very effective. Moreover, the prediction precision of the proposed linear regression model outperforms the existing methods.

关 键 词：查询日志 查询会话 相关查询推荐 线性回归 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]