基于张量分解的跨领域推荐方法

Cross-domain Recommendation System Based on Tensor Decomposition

在海量数据与云计算的背景下,传统的单领域推荐算法很难适应跨领域的信息推荐服务。协同滤波是一种简单常用的推荐算法,但是当目标数据非常稀疏的时候,性能严重退化,借助与目标数据领域关联的辅助数据领域进行跨领域推荐是解决此问题的一种有效途径。已有的跨领域推荐模型大多基于二维评分矩阵,丢失了很多其他维度的信息,导致推荐性能退化。论文提出了一种基于张量分解的跨领域推荐方法,通过提取不同领域的评分模式进行迁移学习填补目标领域空缺值,缓解了数据稀疏性问题,同时提高推荐结果的多样性与准确性。在三个公开的真实数据集上进行的大量实验表明,该模型的推荐精度优于一些目前先进的推荐模型,可适用于大规模信息推荐服务。

In the context of mass data and cloud computing,the traditional single-domain recommendation algorithm is difficult to adapt to cross-domain information recommendation service.Collaborative filtering is a simple and common recommendation algorithm,but when the target domain is very sparse,the performance of serious degradation,with the target domain associated with the field of auxiliary domain for cross-domain recommendations is an effective way to solve this problem.Most of the existing cross-domain recommendation models are based on two-dimensional rating matrix,and many other dimension information is lost,leading to degraded performance.In this paper,a cross-domain recommendation method based on tensor decomposition is proposed,which can reduce the sparseness of data and improve the diversity and accuracy of the proposed results by extracting the learning model in different fields.Many experiments on three publicly presented real data sets show that the model's recommendation accuracy is superior to some of the most advanced recommendation models,it can be applied to large-scale information recommendation service.