基于聚类和SVD++的协同过滤算法

Collaborative Filtering Algorithm Based on Clustering and SVD

奇异值分解(Single Value Decomposition,SVD)++算法是一种基于模型的协同过滤推荐算法,具有良好的推荐效果。然而,随着用户和项目数量的不断增加,用户项目的评分矩阵变得稀疏,导致该算法的推荐结果准确度偏低。为了缓解数据稀疏性问题和用户兴趣随时间漂移问题,本文首先根据用户属性对用户进行聚类操作,其次引入时间因子计算用户间相似度,为用户选取合适的近邻用户,并且根据近邻用户的评分信息为SVD++算法提供了偏差调整项,最后进行了对比实验。实验结果表明,该算法能够准确预测用户评分,提升了推荐效果。

SVD++ algorithm is a model-based collaborative filtering recommendation algorithm, which has good recommendation effect. However, with the increasing number of users and projects, the scoring matrix of user projects becomes sparse, which leads to the low accuracy of SVD++ algorithm recommendation results. In order to alleviate the problem of data sparsity and user interest drifting over time,this paper first clusters users according to user attributes, then introduces time factor to calculate the similarity between users, selects the appropriate nearest neighbor users for users, and provides deviation adjustment items for SVD++ algorithm according to the score information of nearest neighbor users. Finally, a comparative experiment is carried out. Experimental results show that the proposed algorithm can accurately predict user ratings and improve the recommendation effect.