搜索引擎用户访问量模型

基于大规模分布式WWW搜索引擎系统---北大“天网”的用户日志,该文研究了搜索引擎用户访问量建模分析和预测的一般方法;将用户的访问量看成按时间次序排列的随机变量序列,利用时间序列分析的方法,分别建立了天网用户的查询量、点击量和不同IP用户访问量的潜周期模型;结果显示模型对实际数据拟合效果较好;用户访问的主周期为24小时,其它周期依次为12小时、6小时、8小时、5小时、168小时(即一周);用户的异常访问情况可通过小波技术检测。

Modeling Quantity of Users'Access for Search Engine

Tianwang is a large-scale search engine system which is now maintaining index of about 258millions web pages and20millions ftp files.Some hidden periodicity models of the quantity of users' access for search engine is established based on the query and click log of Tianwang.The main idea is that the quantity of users' access,which includes the quantity of users' query,click and distinct IP users under a unit time ,is regarded as a time series.Then the models can be established by applying some methods in time series analysis.The results show that the models provide a better fit to the real data.And the main period of the quantity of users' access is24hour ,the others12hour ,6hour ,8hour ,5hour and168hour (i.e.one week).The exceptional access can be detected by applying wavelet method.