基于K-means聚类与张量分解的社会化标签推荐系统研究

Social tagging recommendation system based on K-means cluster and tensor decomposition

针对大众标注网站推荐系统中存在的数据矩阵稀疏性影响推荐效果的问题,文中采取如下策略:对标注数据进行K-means聚类,将具有相似标签特征的项目进行归类以保证数据具有初始聚合性;聚类完成后运用高阶奇异值分解(high order singular value decomposition,HOSVD)对聚类后的标注数据建立多维张量模型.该策略重点利用张量分解方法对含有用户、标签和项目的三元数据组进行分析,可以进一步改进稀疏性问题,同时形成对项目资源的个性化推荐.通过对社交书签网站Delicious.com的标注数据的处理,验证该方法对解决推荐系统中矩阵稀疏性问题以及提高推荐效果具有改进效果.

Considering the problem of data matrix sparsity which can affect the recommend result on the social tagging recommendation system, we cluster the tagged data whose items have similar tags with K-means method so as to ensure the initial polymerization of these data and use the high order singular value decomposition （ HOS- VD） to build a multidimension tensor model on this database. The point is that the space fensor decomposition has been used to analyse user-tag-item database to solve the matrix sparsity problem and make new personalized recommendation about items to users. According to the processing of data from the social bookmark website - Delicious, com, the new method in this paper has a better result of solving matrix sparsity problem and improves the recommendation.