结合非负矩阵填充及子集划分的协同推荐算法

发布时间：2018-04-29 14:53

  本文选题：低秩矩阵填充 + NMF　； 参考：《小型微型计算机系统》2017年12期

【摘要】：针对协同过滤推荐中评分矩阵极度稀疏问题,以及很多应用对数据存在非负约束要求,提出一种结合矩阵填充及用户-兴趣子集划分的协同推荐算法.首先提出非负约束下的低秩矩阵填充模型(Non-negative Constrained Low Rank Matrix Completion,LR-NM F),以及有效求解该模型的迭代算法.该算法不仅可以利用重构矩阵填充原始矩阵中的缺失项,而且可以得到评分矩阵的非负分解表示.在此基础上,提出一种结合LR-NMF的基于群组的协同推荐方法.利用矩阵非负分解结果,通过块模型近似算法划分用户-兴趣子集或物品-特征子集,最终产生top-N协同推荐列表.实验结果表明,提出的方法不仅有效填充评分矩阵的缺失项,而且推荐精度优于其它协同推荐算法.在大规模稀疏数据集中,仍然具有很好的性能.
[Abstract]:Aiming at the problem of extremely sparse score matrix in collaborative filtering recommendation and the non-negative constraint requirement of many applications, a collaborative recommendation algorithm combining matrix filling and user-interest subset partition is proposed. In this paper, a non-negative Constrained Low Rank Matrix completion model with non-negative constraints is proposed, and an iterative algorithm for solving the model is presented. The algorithm can not only fill the missing items in the original matrix with the reconstruction matrix, but also obtain the non-negative decomposition representation of the score matrix. On this basis, a collaborative recommendation method based on LR-NMF is proposed. By using the matrix nonnegative decomposition result, the block model approximation algorithm is used to divide the user-interest subset or the item-feature subset, and finally the top-N collaborative recommendation list is generated. The experimental results show that the proposed method not only fills the missing items in the scoring matrix effectively, but also has better recommendation accuracy than other collaborative recommendation algorithms. In large sparse datasets, it still has good performance.
【作者单位】： 山东师范大学信息科学与工程学院;山东建筑大学计算机科学与技术学院;
【基金】：国家自然科学基金项目(61672329,61373149,61472233)资助 山东省科技计划项目(2014GGB01617)资助 山东省教育科学规划项目(ZK1437B010)资助 山东省精品课程项目(2012BK294,2013BK399,2013BK402)资助
【分类号】：TP391.3
，

本文编号：1820405