Started in the sparse matrix of recommender systems, and ensured the recommended coverage rate. Based on the I-dea of sociality density, improved the accuracy by matrix reduction. With the user ratings matrix which had been normalized, transformed the consistency score matrix. Meanwhile, with the concept of sociality, got an aggregation analysis result based on user similarity score preferences through adjusting the matrix. Computed the sociality densities by constructing a sociality network diagram. Selected the maximum as the cluster centers, and used the compact and separation effect functions to validate the reliability of aggregation results. Compared to other method, the method reduces the computational complexity. Meanwhile , with the test of Movielens, the method is accuracy and effectiveness in a certain degree.%从推荐系统的矩阵稀疏性问题出发,在保证推荐的覆盖率基础上,借助社区密度的思想并通过矩阵约简来提高推荐的准确率.运用归一化处理的用户评分矩阵对一致性的评分矩阵进行了改造,并借助社区的概念通过矩阵调整得到了一种基于用户相似评分偏好的聚类分析结果；通过构造社区网络结构图得到社区密度,并选择其中的最大值作为聚类中心,运用紧致与分离性效果函数验证了聚类结果的可靠性.该方法的提出,在不损害已有信息的基础上降低了计算复杂度,通过Movielens的数据检验了该方法在一定程度上的正确性和有效性.
展开▼
