一种融合聚类与用户兴趣偏好的协同过滤推荐算法

协同过滤推荐算法可以根据已知用户的偏好预测其可能感兴趣的项目,是现今最为成功、应用最广泛的推荐技术。然而,传统的协同过滤推荐算法受限于数据稀疏性问题,推荐结果较差。目前的协同过滤推荐算法大多只针对用户-项目评分矩阵进行数据分析,忽视了项目属性特征及用户对项目属性特征的偏好。针对上述问题,提出了一种融合聚类和用户兴趣偏好的协同过滤推荐算法。首先根据用户评分矩阵与项目类型信息,构建用户针对项目类型的用户兴趣偏好矩阵；然后利用K-Means算法对项目集进行聚类,并基于用户兴趣偏好矩阵查找待估值项所对应的近邻用户；在此基础上,通过结合项目相似度的加权Slope One算法在每一个项目类簇中对稀疏矩阵进行填充,以缓解数据稀疏性问题；进而基于用户兴趣偏好矩阵对用户进行聚类；最后,面向填充后的评分矩阵,在每一个用户类簇中使用基于用户的协同过滤算法对项目评分进行预测。实验结果表明,所提算法能够有效缓解原始评分矩阵的稀疏性问题,提升算法的推荐质量。

Collaborative Filtering Recommendation Algorithm Combining Clustering and User Preferences

Collaborative filtering recommendation algorithm can use the known user preferences to predict the possible interest items,and it is now the most successful and widely used recommendation technique.However,traditional collaborative filtering recommendation algorithms suffer from data sparsity,which results in poor recommendation accuracy.Only user-item rating matrix has been used to data analysis by most current collaborative filtering algorithms,while the characteristics of the item attributes and user preferences for those who have not been considered.To address this issue,a collaborative filtering recommendation algorithm combing clustering and user preference was proposed in this paper.Firstly,the user preference matrix for the item category is constructed according to the user rating matrix and the item category information.Then the K-Means algorithm is used to cluster the item set,and the nearest-neighbor user corresponding to the unrated item is found based on the user preference matrix.Next,the sparse matrix in each item cluster is filled by the weight Slope One algorithm combined with the similarity of items to alleviate the problem of data sparsity.In addition,the user clusters are built based on the user interest matrix.Finally,the user based collaborative filtering algorithm in each user cluster is employed to predict the item rating for the filled rating matrix.The experimental results show that our algorithm effectively alleviates the sparsity problem of the raw rating matrix and achieves better quality of recommendation than other algorithms.