视频网站中长尾物品的推荐方法

摘要

目前国内外的各大视频网站中均存在严重的长尾效应问题.为降低视频推荐系统中长尾效应的影响,本文提出了基于改进的协同过滤的长尾物品推荐算法PGSim-CF.PGSim-CF算法首先利用pagerank算法计算各物品的初始化权值,然后通过同类视频物品中热门物品的权值去优化冷门物品的权值,与此同时惩罚热门物品的权值,得到各视频物品的最终权值,最后将获得的权值代入协同过滤物品相似度的计算中.实验结果表明,同传统的协同过滤UserCF算法以及文献[1]中所提出的SimCF-ACT算法相比,PGSimCF算法的性能更好.%To address long tail effect in video sites,this paper advanses an improved user-based collaborative filtering algorithm called PGSim-CF.The approach presented in the paper calculates the initialized weight of item based on the pagerank algorithm,and then optimizes the weight of cold item through the weight of popular item in the same category.Finally,the result of the experiment on the MovieLens database show that the proposed algorithm reduce the impact of the long tail effect and have a better performance than traditional collaborative filtering recommendation algorithm and SimCF-AC algorithm.