一种有效缓解协同过滤推荐评价数据稀疏问题的算法  

An Effective Algorithm for Relieving Sparse Data in Collaborative Filtering Recommendation

作　　者：黄永锋[1] 覃罗春[1] 

机构地区：[1]东华大学计算机科学与技术学院,上海201620

出　　处：《东华大学学报（自然科学版）》2013年第1期83-87,共5页Journal of Donghua University(Natural Science)

基　　金：国家自然科学基金资助项目(30770589);中央高校基本科研业务费专项资金资助项目(2011D11206)

摘　　要：在采用协同过滤算法构建个性化推荐的系统中,经常面临用户评价数据稀疏问题,这将严重降低个性化推荐的准确度.针对此问题,提出了一种混合加权预测填充算法,从用户访问的资源特征以及该资源在整个用户群体中被访问的热度出发,对用户访过的但未给出评价的数据进行预测并填充,从而降低了由于用户评价数据缺失所造成的评价矩阵稀疏程度,提高推荐准确度.在MoiveLense数据集上的试验结果表明,该算法能够明显地提高推荐准确度.Personalized recommendation system based on collaborative filtering often faces the data sparsity which seriously reduces the recommendation accuracy. An efficient hybrid weighted prediction algorithm is presented, which predicts the data visited but not rated by the characteristics and the access frequency, and fills the user-item matrix with the predictions. In this way, the sparsity of the user-item matrix is reduced, and the accuracy of recommendation is improved accordingly. Experimental results on MoiveLense data set clearly indicate that the algorithm can significantly improve the recommendation accuracy.

关 键 词：个性化推荐 协同过滤 数据稀疏 预测填充 资源特征 

分 类 号：TP274.5[自动化与计算机技术—检测技术与自动化装置]