摘要
针对电子商务系统中传统协同过滤算法普遍存在的稀疏性和扩展性问题,文中提出了基于模糊C均值聚类有效性的协同过滤算法。首先依据四种不同的聚类有效性函数确定合理的聚类数区间,并在合理聚类数区间中根据Xie-Beni方法搜寻得到最佳的聚类数,然后使用最佳聚类数对项目进行模糊C均值聚类,将用户对单个项目的偏好转化为对相似群组的偏好,将稀疏的用户-项目偏好信息构造成密集的用户-模糊类的偏好信息,最后在项目所属类别中寻找目标用户的最近邻并产生推荐。在数据集Movie Lens上与传统推荐算法相比的实验结果表明,新算法在平均绝对偏差、召回率、准确覆盖率等方面都有了较大改善,提高了推荐质量。
Considering the sparsity and the scalability of traditional collaborative filtering recommendation algorithms in electronic commerce system,a newcollaborative filtering algorithm is presented based on fuzzy C- means clustering validity. Firstly,a reasonable cluster number range is presetted,and then an optimal cluster number is determined based on some representative fuzzy clustering validity functions and Xie- Beni method. Secondly,using the optimal number of cluster,this algorithm transforms the users＇ preferences of single item to similar groups with fuzzy C- means clustering,and sparse user- item preferences is established to dense user- fuzzy preferences.Finally,according to the item ＇s cluster it finds the nearest neighbors of the object user and generates recommendations. The experimental results in M ovie Lens showthat the newalgorithm improves recommendation quality in M AE,recall and coverage.
出处
《计算机技术与发展》
2016年第1期22-26,32,共6页
Computer Technology and Development
基金
国家自然科学基金资助项目(11201290
61104042)
关键词
协同过滤
模糊C均值聚类算法
聚类有效性函数
最佳聚类簇数
collaborative filtering
fuzzy C-means clustering
clustering validity
optimal number of clustering
