摘要
传统的基于余弦相似度度量的云模型协同过滤推荐算法未考虑特征向量的长度和维度,忽略了三个重要数字特征云期望、熵和超熵的关系,如各数字特征具有不同的性质和权重,导致特征丢失、区分度过小的问题。针对这些问题,提出了一种采用标准化的多维欧几里德相似度计算方法,通过将三个数字特征映射为三维空间的点,计算经指数函数标准化的欧几里德相似度,生成更合理的用户k近邻集,最终产生推荐。实验结果表明,该相似度计算方法能够为云特征向量提供更显著的区分度,并在一定程度上提高了推荐质量。
Cosine similarity measurement method is one of the collaborative filtering recommendation algorithms based on cloud model, in which neither the length and dimension of feature vectors nor the relationship among the three digital features of cloud model （cloud expectation, entropy and hyper entropy） are taken into serious consideration. Digital features have different properties and weights, which leads to feature loss and lack of discrimination. Aiming at these problems, we propose a new method which uses Euclidean distance to measure the similarities of the cloud feature vectors. Cloud expectation, entropy and hyper entropy are mapped into the points in a three-dimensional space, and the Euclidean similarities normalized by the exponential function are calculated,thus more proper k-nearest neighbours sets are generated and recommendation results are obtained. Experimental results show that the new similarity measurement method can not only improve the differentiation of cloud feature vectors but also provide a better recommendation quality
出处
《计算机工程与科学》
CSCD
北大核心
2015年第10期1977-1982,共6页
Computer Engineering & Science
基金
国家自然科学基金资助项目(71061008
71462018)
江西省研究生创新专项资金项目(YC2014-S371)
关键词
协同过滤
云模型
数字特征
欧几里德相似度
标准化
collaborative filtering
digital features of cloud model
Euclidean distance similarity
norrealization
