为提高推荐结果的精度和个性化程度,文章有效利用多种信息源,将贝叶 斯方法和深度学习结合,提出一种基于贝叶斯自编码器的社会化推荐算法.算法首先利用混合隶属度随机块模型 MMSB (Mixed membership stochastic block)对用户间交互关系建模,结合用户的属性特征,利用自编码器学习用户的隐含特征向量; 然后利用主题模型结合自编码模块学习物品特征向量; 最后利用概率框架将物品和用户间的各种属性统一起来,共同学习矩阵分解模型中的关系矩阵.模 型中的参数利用变分EM算法进行推理.实验结果表明与同类算法比较,算法在精确度和覆盖率 上有不同程度的提升,且能够得到比较个性化的推荐结果.

In order to improve the accuracy and individuality of the recommendation results, this paper proposes a social recommendation algorithm based on Bayesian self-encoder, which can be achieved by effectively using various information sources to combine Bayesian methods with deep learning. The algorithm in this paper uses the Mixed Membership Stochastic Block Model (MMSB) to model the interaction relationship between users, and uses the self-encoder to learn the user's implicit feature vector combining with the user's attribute features. Item features are learned by combining the LDA model with the self-encoding module. Finally, the relationship matrix in the matrix decomposition model is learned by using probabilistic framework to unify the various attributes between the item and the user. The parameters in the model are sampled by a variational EM algorithm. The experimental results show that compared with similar algorithms, the algorithm proposed in this paper has different degrees of improvement in accuracy and coverage, and can obtain more personalized recommendation results.