中文摘要：

协同过滤技术是推荐系统中核心技术之一，数据的稀疏性和用户的多兴趣性困扰着协同过滤推荐质量的提高。将用户相似性和项目相似性结合起来，对原始评价矩阵进行降维处理，得到对目标评价预测影响最大、数据规模非常小的最近邻评价矩阵，在该矩阵上依照项目近邻程度不同对目标评价预测贡献不同的方法，对用户的邻居进行加权精选，对目标评价实现交错预测。实验结果验证该算法能达到较高的推荐精度。

英文摘要：

Collaborative filtering technology has successfully been applied in many kinds of recommender system, but the issue of sparsity in dataset and multiple-interests of user is an obstacle to quality enhancement of recommendation with collaborative filtering. Taking the similarity of users as well as items into consideration, this paper proposes an algorithm to reduce dimensions of original rating matrix, thus obtaining a nearest-neighbor rating matrix which has small data scale but major influence on predicting target rate. Furthermore, a select method of the nearest neighbors of an active user is presented based on this small matrix according to different contribution of item to target prediction. And cross prediction for target rate is also implemented. The experimental results suggest that this algorithm can provide better recommendation quality.