基于物品属性聚类的融合协同过滤算法.doc

基于物品属性聚类的融合协同过滤算法 //. - 1 - 基于物品属性聚类的融合协同过滤算法 梁佳男，张华** 作者简介：梁佳男（1989-），男，硕士研究生，推荐系统，信息安全 通信联系人：张华（1983-），女，副教授，密码协议、物联网和云计算安全、工业控制系统安全、移动互 联网安全 （北京邮电大学网络与交换国家重点实验室，北京 100876） 5 摘要：随着信息技术的发展，各式各样的推荐系统早已广泛地应用在电子商务、新闻推荐等 领域。协同过滤可以算是推荐系统众多推荐算法中，使用频率最高的一种。但其往往存在数 据稀疏性问题和冷启动等缺点。为了减少推荐系统计算的时间，提高推荐准确率，本文提出 一种基于物品标签聚类和 slope-one 协同过滤的融合推荐算法。在算法中，通过利用用户对 物品的评分行为和物品的属性标签，计算出用户对物品属性的偏好向量。然后利用该向量对10 用户进行聚类，将用户分成多个相似偏好用户组。最后，再使用 slope-one 算法对未评分的 物品进行打分。属性标签的使用，减少了用户特征向量的维度，部分解决了矩阵稀疏的问题， 而通过聚类和 slope-one 算法的融合，在保留了 slope-one 算法计算模型简单优点的同时， 也提高了打分的准确性。最后，通过 MovieLens 上的数据集验证，相比于 slope-one 算法， 该融合算法确实能够提高推荐系统的准确性。 15 关键词：推荐系统；协同过滤；聚类；标签 中图分类号：TP311 Hybrid collaborative filtering algorithm based on Item Tag clustering 20 LIANG Jianan, ZHANG Hua (State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876) Abstract: With the development of information technology, a lot of recommendation systems have been widely used in e-commerce, news recommendation etc. Collaborative filtering is one of 25 the most important recommendation algorithms among recommendation system. However it#039;s also suffering some problems such as data sparsity and cold start. In order to reduce computing time for recommendation and improve the accuracy, a new collaborative filtering recommendation system combining item-tag clustering and slope-one algorithm is proposed. In this algorithm, users were clustered according to users#039; preference on tags of items. Property tags were userd to 30 reduce users#039; feature vector, which solved data sparsity partially. By combining both clustering and slope-one algorithm, the recommendation system retained the advantage of slope-one algorithm while improving accuracy as well. The experiment were applied to MovieLens dataset, which showed that the accuracy of this combined algorithm is in advance of na??ve slope-one.