Abstract: Current neighborhood-based recommendation approaches such as KNN（K-Nearest Neighbor） and K-means collaborative filtering are commonly utilized in predicting user ratings. However, for these methods, the total number K of neighbors is difficult to determine and the recommendation accuracy is limited, which may lead to undesired recommendation effects. Starting from choosing the nearest neighbors, this paper proposes a Collaborative Filtering recommendation integrating user-centric Natural Nearest Neighbor（CF3N） algorithm that self-adaptively searches target user’s natural nearest neighbor set without the parameter K. Then the natural nearest neighbor set is combined with the active nearest neighbor set into an integrated neighbor set, which is exploited to predict target user’s ratings. Experiments are performed to compare the CF3N algorithm with CF-KNN and INS-CF algorithms using MovieLens datasets with the Mean Absolute Error（MAE） and the Root Mean Square Error（RMSE） as evaluation indexes. The results show that the CF3N algorithm significantly improves the recommendation accuracy, thus providing a new approach to predicting user ratings in movies.

Key words: recommender system, collaborative filtering, natural nearest neighbor, parameter-free, recommendation accuracy

摘要： 现有的基于近邻的协同过滤推荐方法如基于KNN、基于K-means的协同过滤推荐常用来预测用户评分，但该方法确定邻居个数K非常困难且推荐准确率不高，难以达到理想推荐效果。从选择邻居用户这一角度出发，提出一种融合用户自然最近邻的协同过滤推荐算法（Collaborative Filtering recommendation integrating user-centric Natural Nearest Neighbor，CF3N），该算法首先自适应地寻找目标用户的自然最近邻居集，再融合目标用户的自然最近邻居集与活动近邻用户集，使用融合后得到的邻居集合预测目标用户评分。实验使用了MovieLens数据集，以RMSE和MAE为评测标准，比较CF3N、CF-KNN与INS-CF算法，结果显示在电影领域该算法的推荐准确率有显著提高。

关键词: 推荐系统, 协同过滤, 自然最近邻用户, 无参, 推荐准确率