基于时间的局部低秩张量分解的协同过滤推荐算法

传统的推荐模型是静态的,忽略了时间因素。部分推荐算法虽然将时间因素考虑在内,但只是简单使用最近的数据或者 降低 过去数据的权重,这样可能会造成有用信息的丢失。针对这一问题,提出了一种考虑时间因素的局部低秩张量分解推荐算法。在传统的推荐算法的基础上,放松用户对项目的评分矩阵是低秩的这一假设,认为整个评分矩阵可能不是低秩的而是局部低秩的,即特定用户项目序偶的近邻空间是低秩的；同时又考虑时间因素,把评分矩阵看作是用户、项目和时间3个维度的张量,将传统的推荐算法延伸到张量领域。实验表明,所提算法能显著提升排名推荐性能。

Time-based Local Collaborative Filtering Recommendation Algorithm on Tensor Factorization

Traditional recommendation models are stationary with neglecting time factor.Some recommendation algorithms take time factor into consideration,but what they do is using the latest data or reducing the weight of past data.It may lead to the loss of some useful information.To solve the above problem,a time-based local low-rank tensor factorization algorithm was proposed.In contract to standard collaborative filtering algorithms,our method does not assume that the rating matrix is low-rank.We relaxed the assumption and assumed that the rating matrix is locally low-rank.The algorithm takes time factor into consideration and views rating matrix as 3-dimensional sensor based on the traditional recommendation algorithms which extend the traditional algorithms to tensor field.Experiments show that the algorithm could improve the efficiency of ranking recommendation.