摘要
针对协同过滤算法中存在的数据稀疏性、可扩展性及准确性问题,提出一种基于改进矩阵分解和谱聚类的协同过滤算法。该算法首先将通过抑制物品流行度和用户活跃度优化的相似度计算融入最小二乘法(ALS),以避免矩阵分解时因子信息的丢失;其次结合流形学习的谱聚类算法弥补ALS算法产生的大计算量问题,同时获得全局最优解以提高聚类所得目标用户最近邻居的准确率;最后利用Movielens数据集进行实验。实验结果表明,改进的算法可以有效降低协同过滤算法的平均绝对误差和均方根误差,提高准确率,拥有更优的性能。
A collaborative filtering algorithm based on improved matrix factorization and spectral clustering is proposed to address the issues of data sparsity,scalability,and accuracy in collaborative filtering algorithms.The algorithm first incorporates similarity calculation optimized by suppressing item popularity and user activity into the least squares method(ALS)to avoid the loss of factor information during matrix decomposition.Secondly,manifold learning algorithm based on spectral clustering is used to compensate for the high computational complexity caused by the ALS algorithm,while obtaining the global optimal solution to improve the accuracy of clustering the nearest neighbors of the target user.Finally,experiments are conducted using the Movielens dataset.The experimental results show that the improved algorithm can effectively reduce the average absolute error and root mean square error of the collaborative filtering algorithm,improve accuracy,and have better performance.
作者
舒珏淋
谢红韬
袁公萍
SHU Juelin;XIE Hongtao;YUAN Gongping(CETC Big Data Research Institute Co.,Ltd.,Guiyang 550002,China)
出处
《现代信息科技》
2024年第9期73-76,共4页
Modern Information Technology
基金
国家自然科学基金(U19B2027)。
关键词
协同过滤算法
相似度
谱聚类
全局最优解
collaborative filtering algorithm
similarity
spectral clustering
global optimal solution
作者简介
舒珏淋(1992-),男,侗族,贵州镇远人,硕士在读,研究方向:数据挖掘。
