Abstract: With the fast development of the Internet, there is high demand for three-dimension (3D) recommendation services in the fields of personalized Web search, personalized advertisement delivery and personalized social bookmarks. It is well known that the cube data are highly sparse and the relationships of 3D objects in the 3D collaborative filtering recommendation system are more complex than those in the 2D collaborative filtering recommendation system. In order to better model the relationships of 3D objects and address the sparsity problem of the cube data, this paper proposes a novel approach to 3D collaborative filtering recommendation, referred to as cube alternating least squares (CubeALS). CubeALS improves the 3D collaborative filtering recommendation algorithm, cube singular value decomposition (CubeSVD), by applying ALS algorithm in the 2D matrix decomposition. The experimental evaluation using a real-world dataset shows that CubeALS achieves better results in comparison with CubeSVD.

Key words: cube alternating least squares (CubeALS), cube singular value decomposition (CubeSVD), 3D collaborative fil-tering, personalized recommendation

摘要： 随着互联网的快速发展, 人们对个性化网页搜索、个性化广告投放、个性化社会标注等三维推荐服务的需求越来越紧迫。这些三维立方体数据高度稀疏, 且与二维推荐系统相比三维推荐系统中对象之间的关系更加复杂。为了更好地模拟三维对象之间的关系并解决三维数据高度稀疏的问题, 提出了一种新的三维协同过滤推荐算法CubeALS(cube alternating least squares)。该算法对三维协同过滤推荐算法CubeSVD (cube singular value decomposition)进行了改进, 尝试使用不同于SVD的算法进行矩阵分解。在真实的个性化社会标注数据集上的实验结果表明, 与CubeSVD算法相比, CubeALS的性能得到了显著提高。

关键词: CubeALS, CubeSVD, 三维协同过滤, 个性化推荐