基于用户特征迁移的协同过滤推荐

Collaborative Filtering Recommendation Based on User Feature Transfer

为提高推荐系统在数据稀疏情况下的推荐质量,提出一种基于用户特征迁移的协同过滤推荐模型。利用矩阵分解技术提取辅助领域的用户特征,通过建立正则项约束的矩阵分解模型,将辅助领域的用户特征迁移到目标领域中,协助目标领域用户特征的学习,最终生成目标领域的用户推荐。设计快速收敛的Wiberg算法得到模型的最优解,并对实际应用中的可行性进行分析。通过对2个公开数据集的实验结果表明,该模型能够实现辅助领域用户特征的迁移,有效提高目标领域的推荐质量。

In order to improve the recommendation quality of recommender system with data sparsity,this paper proposes a user collaborative filtering recommendation model based on feature transfer. Firstly,matrix factorization technology is applied to collect the user feature from the auxiliary domain. Secondly,it constructs a matrix factorization model with the constraint of regularization term,which is used to transfer the user feature learned from the auxiliary domain to the target domain,so as to help the learning of user feature in the target domain. Finally,user recommendation is made for the target domain. A fast convergence Wiberg algorithm is also designed for the model to get the optimal solution,whose feasibility is also discussed for practical application. Through the experiment on two real world data sets,the model can effectively transfer the user feature of source domain,and improve the quality of recommender system for target domain.