Abstract: In terms of the problems that the collaborative filtering algorithm used in traditional user personalized recommendation has sparseness and lacks of scalability, this paper proposes a collaborative filtering algorithm based on user features clustering and Slope One filling. In the algorithm, firstly, the user attribute feature is used as the clustering basis, and the K-means clustering algorithm based on minimum spanning tree is used to cluster the users and generate K similar user sets. Secondly, based on the clustering analysis, the Slope One algorithm is used to predict and fill the user score matrix under the similar user sets. Finally, a hybrid cooperative filtering algorithm is used to perform the nearest neighbor search on the filled user score matrix to obtain the predicted score and produce the recommended result. The comparative experimental results show that the proposed algorithm can significantly improve the accuracy of the recommendation, effectively alleviate to the sparseness problem and have good expansibility.

Key words: collaborative filtering, user characteristics, K-means, Slope One, nearest neighbor search

摘要： 针对传统的用户个性化推荐中使用的协同过滤算法存在稀疏性和可扩展性不足的问题，提出了一种基于用户特征聚类和Slope One填充的协同过滤算法。该算法首先以用户属性特征作为聚类依据，利用基于最小生成树K-means聚类算法对用户进行聚类分析，生成K个相似用户集合；其次在聚类分析的基础上，利用Slope One算法预测填充生成的相似用户集下的用户评分矩阵；最后采用混合协同过滤算法对填充后的用户评分矩阵进行最近邻搜索，从而得到预测评分，产生推荐结果。对比实验结果表明，提出的算法显著提高了推荐的精度，有效缓解了稀疏性问题，具有良好的可扩展性。

关键词: 协同过滤, 用户特征, K-means, Slope One, 最近邻搜索