摘要
协同过滤算法是电子商务推荐系统中最重要的技术之一,基于传统的协同过滤推荐技术并未考虑到新商品有更多的推荐价值。提出一种改进策略,运用矩阵分解SVD算法、余弦相似性,将具有共同兴趣的用户聚簇分组,提取组内用户评价产品的特征向量,运用BP神经网络进行训练,预测用户组对未知产品的满意度。对于满意的新产品赋予较高的推荐等级,进行优先推荐。同时,也有效弥补了在传统的协同过滤算法中普遍存在着的冷启动、矩阵稀疏等问题。
Collaborative filtering algorithm is one of the most important technologies in e-commerce recommendation system, but tradition-based collaborative filtering recommendation technology does not take into account the new goods that have more recommendation values. In this paper we propose an improved strategy, which uses matrix decomposition SVD algorithm and cosine similarity to group users clustering with common interests and to extract the eigenvector of products to be evaluated by the users in group. By using BP neural network for training, it predicts the satisfaction of users group on unknown products. For those satisfied new products it assigns higher recommending grade, and gives the priority to recommending them. At the same time, the strategy effectively makes up the problems of cold start and sparse matrix, etc. commonly existed in traditional collaborative filtering algorithms.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第10期285-287,328,共4页
Computer Applications and Software
基金
宁夏自然科学基金重点项目(NZ13004)
关键词
协同过滤
矩阵分解SVD
余弦相似性
BP神经网络
Collaborative filtering
Matrix decomposition SVD
Cosine similarity
BP neural networks
