72-罗辛-通过相似度支持度优化基于K近邻的协同过滤算法.pdfVIP

通过相似度支持度优化基于 K 近邻的协同过滤算法 罗辛 欧阳元新 熊璋 袁满 （北京航空航天大学计算机学院，北京，100191） 摘 要 个性化推荐系统能基于用户个人兴趣为用户提供定制信息。此类系统通常使用协同过 滤技术实现，其中一种广泛使用的经典模型是基于用户评分相似度的K 近邻模型。使用K 近邻 模型需要预先计算出用户或者项目的K 个最近邻居，K 值过大时会导致计算量过大而影响推荐 产生的实时性，而K 值过小则会导致推荐精度下降。为解决此问题，本文中提出了一种新的最 近邻度量——相似度支持度。基于相似度支持度，本文提出了数种能够在保持推荐精度和密度 的前提下维持合理规模的K 近邻的策略。在真实大规模数据集上的实验结果表明，相比传统算 法，本文提出的策略能够在保证推荐精度的前提下大幅降低计算复杂度。 关键词 个性化推荐；协同过滤 中图法分类号 TP 391.4 The Effect of Similarity Support in K-nearest-neighborhood Based Collaborative Filtering Xin LUO Yuanxin OUYANG Zhang XIONG Man YUAN (School of Computer Science, BeiHang University, Beijing 100191) Abstract Recommender systems which can provide people with personalized suggestions usually rely on Collaborative Filtering (CF). A classical approach to CF is based on K-nearest-neighborhood (KNN) model, where the most important task is constructing the KNN sets for involved users or items. However, when constructing KNN sets, there is a dilemma to decide the value of K--a too small value will lead to poor recommendation performance, whereas a too large one will result in unacceptable computational complexity. In this work we first empirically validated that the suitable value of K in KNN based CF was affected by the number of the totally involved entities, and then focused on improving the quality of the KNN sets in KNN based CF for providing high recommendation performance as well as maintaining suitable KNN set size. To achieve this objective, we proposed a novel KNN metric named Similarity Support (SS). By taking SS into consideration during the KNN building process, we designed a series of strategies for optimizing KNN based