P2P网络资源搜索中的用户行为模型  

User behavior model for search in peer-to-peer networks

作　　者：钱宁[1,2] 胡永东[1,2] 吴国新[1,2] 

机构地区：[1]东南大学计算机网络和信息集成教育部重点实验室,南京210096 [2]东南大学计算机科学与工程学院,南京210096

出　　处：《东南大学学报（自然科学版）》2011年第4期706-710,共5页Journal of Southeast University：Natural Science Edition

基　　金：国家高技术研究发展计划(863计划)资助项目(2007AA01Z422)

摘　　要：为了改善P2P网络的搜索性能,提出应用马尔可夫链预测用户兴趣集合的方法.根据用户的历史查询行为,对关键词进行聚类,得到相关的兴趣类;应用马尔可夫链,建立用户行为模型,描述用户兴趣的时序变化;根据极限概率,预测用户达到平稳状态时的兴趣集合;给出根据模型进行预测的方法.实验表明,模型的命中率受查询序列分布的影响较大,增加兴趣集合规模可以提高模型的性能,模型的性能还受到观察长度的影响.当节点表现出明显的兴趣特征时,用户行为模型具有较高的预测准确度.To improve search performance in P2P networks,a novel method based on Markov chain is proposed to predict interests of peers.Firstly,based on past queries,Key words are clustered and interests related to peers are obtained.Then,a user behavior model is established using Markov chain to describe the changes of user＇s interests over time.Finally,limiting probability is employed to predict user＇s interests in the future.The predicting method based on this model is also presented.Experimental results demonstrate that the performance of the model is influenced by the sequence of query.Increasing the size of interest set can improve the performance.The performance is also affected by observing length.This approach can achieve high hit rate under various settings when peers show obvious preference.

关 键 词：P2P网络 用户行为 马尔可夫链 极限概率 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]