摘要
针对传统单一推荐算法难以兼顾用户冷启动、数据高维稀疏、算法准确性与可扩展性等方面的问题,提出一种基于多源数据聚类和奇异值分解的混合推荐算法。该算法首先利用TF-IDF公式对用户—项目评分矩阵和项目特征矩阵进行处理,生成用户—项目偏好矩阵;然后结合用户特征矩阵、评分矩阵作为算法输入,利用改进K-means聚类算法划分用户类簇;接着采用融合时间衰减函数的BiasSVD算法对每个用户类簇对应的评分矩阵进行分解降维,并利用随机梯度下降法重新预测评分填充评分矩阵;最后从高到低对用户的预测评分向量进行排序,产生推荐列表。在MovieLens数据集上的实验结果证明,该算法的精确率和召回率相较于传统基于SVD的协同过滤推荐算法分别提升5.4%和6.8%,表现出更好的准确性和推荐性能,改善了用户冷启动问题。所提方法对目前混合推荐算法具有一定的参考与借鉴价值。
A hybrid recommendation algorithm based on multi-source data clustering and singular value decomposition is proposed to solve the problems of traditional single recommendation algorithm,such as user cold start,high-dimensional sparse data,algorithm accuracy and scalability.Firstly,the algorithm uses TF-IDF formula to process the user item scoring matrix and item characteristic matrix,and generates the user item preference matrix;Second,combined with the user characteristic matrix and scoring matrix as the algorithm input,the improved k-means clustering algorithm is used to divide user clusters;Third,the bissvd algorithm with time decay function is used to decompose and re⁃duce the dimension of the score matrix corresponding to each user class cluster,and the random gradient descent method is used to re predict the score and fill the score matrix;Finally,the user′s prediction score vector is sorted from high to low to generate a recommendation list.The experimental results on movielens dataset show that the accuracy and recall of this algorithm are improved by 5.4%and 6.8%respectively com⁃pared with the traditional collaborative filtering recommendation algorithm based on SVD,showing better accuracy and recommendation perfor⁃mance,and improving the cold start problem of users.The proposed method has a certain reference for the current hybrid recommendation algorithm.
作者
刘伟友
吴陈
LIU Wei-you;WU Chen(College of Computer Science,Jiangsu University of Science and Technology,Zhenjiang 212100,China)
出处
《软件导刊》
2022年第11期75-81,共7页
Software Guide
关键词
混合推荐
聚类
奇异值分解
多源数据
协同过滤
推荐算法
hybrid recommendation
clustering
singular value decomposition
multi-source data
collaborative filtering
recommended algorithm
