杰卡德相似系数在推荐系统中的应用  

Application of Jaccard Similarity Coefficient in Recommender System

作　　者：张晓琳[1] 付英姿[1] 褚培肖 

机构地区：[1]昆明理工大学理学院,云南昆明650500

出　　处：《计算机技术与发展》2015年第4期158-161,165,共5页Computer Technology and Development

基　　金：国家自然科学基金资助项目(11201200)

摘　　要：项目相似性度量是协同过滤系统的核心。相关研究中,基于物品协同过滤系统的相似性度量方法普遍使用余弦相似性。然而,在许多实际应用中,评价数据稀疏度过高,物品之间通过余弦相似度计算会产生误导性结果。文中将杰卡德相似性度量应用到基于物品的协同过滤系统中,并建立起相应的评价分析方法。与传统相似性度量方法相比,杰卡德方法完善了余弦相似性只考虑用户评分而忽略了其他信息量的弊端,特别适合于文中所应用的稀疏度过高的数据。最后通过实例说明上述方法的有效性。Item similarity metric is the core of collaborative filtering system. In related research, cosine similarity has been regularly used in item-based collaborative filtering system. However, in many practical cases, cosine similarity leads to misleading results due to high sparsity. Jaccard similarity is used in item-based collaborative filtering in this article to build the corresponding evaluation and analysis methods. Compared with traditional similarity metric, Jaccard method improves the drawbacks of cosine similarity, which only takes user rating in consideration and ignores other information ,and is suitable for considering the high sparse data. In the end,an example is used to prove the effectiveness of Jaccard similarity.

关 键 词：推荐系统 协同过滤 杰卡德相似系数 余弦相似性 

分 类 号：TP302.1[自动化与计算机技术—计算机系统结构]