摘要
个性化推荐系统是电子商务系统的一个重要研究内容,计算顾客之间的相似性或顾客聚类是产生良好推荐的关键．通过分析个性化推荐的应用特征,即顾客评分数据稀疏及其影响；在开放的电子商务环境中,新顾客不断加入和顾客偏好的迁移,使顾客簇不断发生变化,提出了一种基于自组织特征映射聚类的协同过滤推荐算法,对高堆稀疏的样本进行动态聚类．它具有下列特点：①在自组织特征映射聚类中,引入抑制函数,使其能够适应顾客评分数据的稀疏性；②设置神经元的分裂和合并过程,使其能够满足顾客聚类的动态变化．通过实验分析,表明该算法能够适应顾客评分数据稀疏和顾客聚类的动态变化特征,从而提高推荐质量．
Personalized recommender system becomes an important research field in electronic commerce （EC）, and the clustering of customers is the basis to produce the recommendation. The characters of customers＇ clustering in EC are distinct from other applications, such as the extreme sparsity of user rating data, the frequent alteration of clustering of customers because of transference of customer＇ s performance and more and more new customers. So, traditional methods work poor in the situation. To address those issues, a novel collaborative filtering algorithm based on self-organizing feature maps is proposed. The model has following features ： ① a restraint function is introduced in the basic model of SOM, to solve clustering of sparse data;② the splitting process and merging process of neuron are constructed, to realize dynamic clustering in EC environment. The experimental results show that the model can efficiently improve the quality of clustering of customers, and make better recommendation.
出处
《天津大学学报》
EI
CAS
CSCD
北大核心
2006年第B06期279-282,共4页
Journal of Tianjin University(Science and Technology)
基金
国家自然科学基金(70402009)
天津市高校科技发展基金(20030613)
天津师范大学博士基金(52RL005).
关键词
协同过滤
推荐算法
聚类分析
自组织特征映射
collaborative filtering
recommendation algorithm
clustering
self-organizing feature maps
