基于填充先验约束的矩阵分解算法

Imputed prior constraint⁃based matrix factorization

尽管现有的社会推荐方法,特别是基于矩阵分解的社会推荐方法,取得了一定的推荐效果,但这些方法使用评分数据空间的用户偏好去约束社交关系数据空间的用户偏好,而这两种用户偏好却处于不同的数据空间,这限制了推荐模型的准确性.为解决这个问题,提出一种使用填充数据的偏好来约束评分数据偏好的学习过程的方法.该方法首先设计一个算法生成填充数据,然后,在概率矩阵分解的过程中约束填充数据偏好的先验分布服从评分数据偏好的先验分布.在四个真实数据集(TrustFilm,Ciao,MovieLens 1m和Jester)上测试的结果表明,提出方法的推荐效果比现有的代表性方法都要好,为概率矩阵分解模型中先验约束的研究提供了新思路.

Although existing social recommendation methods,especially the social recommendation methods based on matrix factorization,achieve good recommendation performance,these methods use user preferences of rating data to constrain that of social data,which are from different data spaces.This procedure limits the accuracy of recommendation model.To solve this problem,the learning approach to constrain the preferences of rating data by using the user preferences of the imputed data is proposed.The proposed method designs an algorithm to generate imputed data firstly,and then constrains the prior distribution of the preferences from the imputed data to obey the distribution of the prior distribution of the preferences from the rating data.We evaluate our method on four real⁃life datasets,i.e.,TrustFilm,Ciao,MovieLens 1m and Jester.The experimental results show that our method outperforms the state⁃of⁃the⁃art methods in recommendation accuracy.This work provides new idea for the research on the prior constraint in the model of probabilistic matrix factorization.