基于流形排序的社会化推荐方法

提出一种基于流形排序和社会化矩阵分解的推荐方法，采用流形排序方法度量用户间的社会相似度，利用正则化技术构建用于评分矩阵因式分解的目标函数，将用户之间的偏好差异作为目标函数的惩罚项，从而将用户之间的社会相似性融入评分矩阵的低阶矩阵分解过程. 实验结果表明，在大型的数据集上，该方法获得了比当前同类方法更好的推荐精度和更低的评分预测均方根误差/评分预测平均绝对误差（RMSE/MAE）值.

Social Recommendation Based on Manifold Ranking

A new recommendation method based on manifold ranking and social matrix factorization is proposed, in which the social similarities among users are calculated by means of manifold ranking, the objective function of ratings matrix factorization is constructed via the regularization technique, with the differences among users' preferences as the penalty of objective function, the social similarities are infused into the low-rank matrix factorization. Experiments show that this method achieves higher precisions and lower root mean square error/mean absolute error (RMSE/MAE) value than other that of cognate methods.