时间加权不确定近邻协同过滤算法  

Time-weighted Uncertain Nearest Neighbor Collaborative Filtering Algorithm

作　　者：郑志高[1] 刘京[2] 王平[3] 孙圣力[1] 

机构地区：[1]北京大学软件与微电子学院,北京100260 [2]北京大学信息科学技术学院,北京100871 [3]北京大学软件工程国家工程研究中心,北京100871

出　　处：《计算机科学》2014年第8期7-12,共6页Computer Science

基　　金：江苏省自然科学基金项目(BK2010139)资助

摘　　要：围绕传统的协同过滤推荐算法存在的局限性展开研究,提出一种时间加权不确定近邻协同过滤推荐算法TWUNCF。根据推荐系统应用的实际情况,首先对用户和产品相似度进行时间加权以保证数据有效性,在此基础上改进相似度的计算方法。同时引入近邻因子在产品群和用户群中自适应地选择预测目标的近邻对象作为推荐群,计算推荐群中推荐概率较高的信任子群,最后通过不确定近邻的动态度量方法来对预测结果进行平衡的推荐。实验结果表明,该算法考虑了数据的时间有效性,同时平衡不同群体对推荐结果的影响,避免由于数据稀疏带来的推荐结果不准确和计算难度大的问题。理论分析和模拟实验证明,该算法在一定程度上提高了系统的准确性和推荐效率。To overcome the limitations of the traditional collaborative filtering recommendation algorithm, this paper proposed a Time-Weighted Uncertain Nearest Neigbbor Collaborative Filtering Algorithm （TWUNCF）. According to the actual application situation of recommendation system, the author weighted the product similarity and user similarity to ensure the data validity firstly,and then improved the calculation method of the similarity. And then the author intro- duced the neighbor factor to select the trusted neighbors of the recommendation object adaptively. Based on these, bal- anced the prediction result by using dynamic metrics of uncertain nearest neighbors. Experimental results show that the algorithm can be used to improve data validity according to the time attribute, and balance the impact of the different groups on the recommendation result, and avoid the problems caused by the data sparseness. Theoretical analysis and simulation experiment show that the algorithm this paper proposed outperforms existing algorithms in recommendation quality, and improves the system＇s accuracy and recommendation efficiency.

关 键 词：协同过滤算法 时间权重 不确定近邻 信任子群 推荐系统 

分 类 号：TP274.2[自动化与计算机技术—检测技术与自动化装置]