基于Hash的Top-N推荐方法  

Top-N recommender method based on Hash strategy

作　　者：朱婷[1] 陈德 唐长兵 郑忠龙[1] 

机构地区：[1]浙江师范大学数理与信息工程学院,浙江金华321004

出　　处：《浙江师范大学学报（自然科学版）》2018年第1期56-63,共8页Journal of Zhejiang Normal University:Natural Sciences

基　　金：国家自然科学基金资助项目(61672467;61272007);浙江省自然科学基金资助项目(LY14F030008;2015C31095)

摘　　要：针对电子商务数据量大、用户寻找有用信息困难的现状,提出了基于Hash的Top-N推荐方法.通过两步骤Hash策略,并利用主成分分析(PCA)法,将数据降维后再通过k-means聚类量化;然后运用协同过滤,以二进制码对应实值的Manhattan距离度量用户相似性;最后计算推荐项的预测评分,将推荐列表中的前N项作为最终的推荐项目呈现给用户.结果表明:命中率(HR)与平均命中等级倒数(ARHR)的结果较好,该方法能够有效地进行个性化推荐.For today' s large volume e-commerce data,and difficulties from the users while finding useful information,it was proposed Top-N recommended method based on Hash strategy. By adopting the two stepHash strategy,using the principal component analysis(PCA) to reduce the dimension of dataset,quantifying the dataset,using collaborative filtering to calculate the user and item similarity and prediction score,the top N items were selected and presented to the recommended list as final recommendations for the users. The result showed that the effect of HR and ARHR was better,the proposed method was effective for personalized recommendation.

关 键 词：Hash学习 主成分分析 推荐系统 协同过滤 K-MEANS聚类 Manhattan距离 

分 类 号：TP302[自动化与计算机技术—计算机系统结构]