应用非负矩阵分解的社交网络好友推荐

现有好友推荐方法只利用用户关系或内容信息进行推荐，难以获得较好的推荐质量． 针对该问题，在利用非负矩阵分解模型适合数据聚类以及数据约简的基础上，提出一种基于非负矩阵分解的好友推荐方法:FRNMF． 该方法采用基于非负矩阵分解的用户聚类为核心的好友推荐框架，利用用户好友关系网络信息和内容信息分别进行用户聚类，然后基于聚类结果计算用户间的综合相似度并进行好友推荐；不仅可以综合集成利用用户关系和内容两类信息，而且具有线性时间复杂度，还可以解决数据稀疏引起的推荐质量下降问题． 实验开发了FRNMF的原型系统，并在真实的新浪微博和学者网社交网络数据集进行对比实验，结果表明FRNMF比传统的好友推荐方法具有更好的推荐质量． 此外，对用户关系和内容两类信息的权重参数设置进行实验分析，分析表明适当提高用户关系信息的权重对于提高好友推荐质量具有促进作用．

Friend Recommendation in Social Network Using Nonnegative Matrix Factorization

Most of existing friend recommendation methods only utilize user friendship or content information, and hence they are hard to obtain better recommendation quality. Aiming at this problem, Friend Recommendation using Nonnegative Matrix Factorization (FRNMF) for friend recommendation based on Nonnegative Matrix Factorization (NMF) is proposed, which is fit for data clustering and data reduction. FRNMF adopts user clusters as the core component of its framework. It firstly clusters users by utilizing user friendship network and user-generated content information respectively, and then calculates user pairwise similarities for recommendation based on the cluster results. It can use both user friendship and content information, and it has linear time complexity. FRNMF can alleviate the problem of data sparsity, which can result in the low recommendation quality. By developing protosystem of FRNMF and conducting comparative experiments on Weibo and Scholat social networks, the results show that our method performs better than traditional friend recommendation methods. Moreover, by experimental analysis, moderate increase of the weight of user friendship information can further improve the recommendation quality.