CAJ | 학술논문

提出了基于联合聚类和带正则化的迭代最小二乘法的协同过滤算法。该算法对原始矩阵进行用户-项目两个维度的联合聚类生成若干子矩阵，子矩阵的规模远小于原始评分矩阵，可有效降低预测阶段计算量，而且也缓解了数据稀疏性问题。在子矩阵中通过对传统的矩阵分解进行正则化约束来防止模型过拟合现象，并采用迭代最小二乘法进行训练分解模型，可有效缓解可扩展性。实验表明，该方法具有高效性。
제출료기우연합취류화대정칙화적질대최소이승법적협동과려산법。해산법대원시구진진행용호-항목량개유도적연합취류생성약간자구진，자구진적규모원소우원시평분구진，가유효강저예측계단계산량，이차야완해료수거희소성문제。재자구진중통과대전통적구진분해진행정칙화약속래방지모형과의합현상，병채용질대최소이승법진행훈련분해모형，가유효완해가확전성。실험표명，해방법구유고효성。
This paper proposes a collaborative filtering algorithm based on co-clustering and alternating-least-squares with weighted-regularization .The algorithm divides the original matrix into several sub-matrix,and the sub-matrix is much smaller than the size of the original scoring matrix , which not only reduces the amount of computation , but also alleviates the problem of data sparsity .In the sub-matrix by using regularization constraint to prevent model from over fitting and by using least-squares method to train decomposition model ,the scalability can be effectively alleviated .The experiments show that this method is efficient .