基于用户相似性选择及标签距离的推荐算法

Recommendation Algorithm Based on User Similarity Selection and Label Distance

邻居选择和物品的标签信息对于推荐系统进行评分预测具有重要的影响。为了解决推荐系统中存在的预测和排序准确性低及算法可拓展性不足的问题。本文基于距离模型算法,提出了基于用户相似性选择及标签距离的推荐算法。首先,选取相似性大于阈值的用户作为待预测用户的邻居来应对算法可拓展性不足的问题。其次,利用物品的标签信息,将用户对物品的评分映射为用户对物品标签的评分以提高准确性。通过在两个电影数据集中利用十折验证预测用户对电影的评分。实验结果表明,基于用户相似性选择及标签距离的推荐算法在准确性以及可拓展性中均获得了较大提升。

Neighbor selection and item label information have important influence on rating prediction of recommendation system.To improve the accuracy and scalability of recommendation systems,this paper proposes a distance-model based approach that utilizes user similarity selection and label distance.First,the users with similarity greater than the threshold value are selected as the neighbors of the users to be predicted to deal with insufficient scalability of the algorithm.Second,the user’s rating of the item is mapped to the user’s rating of the item label using the label information to enhance the accuracy.Users’ratings of movies were predicted by using discount validation in both movie datasets.Experimental results show that the accuracy and scalability of the recommendation algorithm based on user similarity selection and label distance are greatly improved.