中文摘要：

现有的知识推荐方法主要是静态知识分类推荐和单个用户个性化推荐,忽略了用户大众在知识访问中表现出的网络集群行为特征。用户的网络集群行为所访问的知识项之间往往隐含着某些内在联系,将网络集群行为下所形成的知识群落称为知识簇。根据基于网络集群行为的用户访问之间的影响关系和访问日志处理过程,构建了基于网络集群行为的动态知识簇模型。将用户访问的知识项看作网络节点,利用概率推理得出节点之间的关联关系形成动态知识簇。当用户访问某个知识项时,根据动态知识簇向用户推荐该知识项的相关知识。使用网络爬虫技术挖掘知识服务网＂豆瓣网＂用户对豆瓣电影的访问日志作为实验数据,实验结果证明了基于概率推理的动态知识簇的推荐方法是有效的。

英文摘要：

The existing knowledge recommendation methods mainly based on static knowledge classification recommendation and personalized knowledge recommendation,which ignores the collective behavior characteristic of massive users accessing knowledge items.The concept of knowledge cluster is proposed according to the implicit relations between knowledge items accessed by internet collective behavior of massive users.A dynamic model of knowledge cluster based on internet collective behavior is put forward by the influence of user access in internet collective behavior and processing procedure of access logs.Accessed knowledge items are regarded as nodes in probabilistic reasoning which forms the dynamic knowledge cluster.When users access a knowledge iterm,relevant knowledge iterms in the knowledge cluster are recommended.Experiment data is collected from access logs in knowledge service site ＂douban＂ by Web Crawlers,and the experiment result of dynamic knowledge cluster based on probabilistic reasoning is proved to be valid.