一种基于聚类预填充的SVD算法

A SVD Algorithm Based on Clustering Prefilling

在大多数推荐系统中,由于用户评分数据的稀疏性,使得相似度误差较大并影响推荐结果的准确性。为了解决用户评分数据的稀疏性问题,提出了一种基于聚类预填充的SVD算法,依据项目的特征和用户对项目的评分,生成用户项目—特征兴趣矩阵,使用用户项目—特征兴趣矩阵对用户进行聚类,将同类用户对项目的平均评分作为未评分项目的估计评分填充用户—项目评分矩阵,再使用SVD算法对未评分项目预测评分并产生推荐结果。通过实验证明,在选取不同最近邻数量的条件下,提出的基于聚类预填充的SVD算法相比传统算法,推荐精度和准确度有一定程度的提高。

In most recommendation systems,sparseness of user rating data causes a large deviation and affects recommendation results.In order to solve the problem,this paper proposes a cluster pre-filling SVD(Singular Value Decomposition)algorithm.The algorithm generates user-item preference matrix based on the characteristics of the items and the user’s rating,then clusters users,and fills the user-item rating matrix with the average score of similar users on the item,then uses the SVD algorithm to predict the score of unrated items and produce recommended results.It is experimentally proven that,compared with the traditional collaborative filtering recommendation algorithm and SVD algorithm,the cluster pre-filling SVD algorithm precision and accuracy have been improved to a certain degree.