考虑负相关性信息的协同过滤算法研究  

Collaborative filtering algorithm by considering negative correlations

作　　者：郭强[1] 周继平[1] 郭迎迎[1] 胡兆龙[1] 刘建国[1] 

机构地区：[1]上海理工大学复杂系统科学研究中心,上海200093

出　　处：《计算机应用研究》2013年第12期3543-3545,3575,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(91024026;71071098;71171136);上海市科研创新基金资助项目(11ZZ135;11YZ110);国家教育部科学技术研究重点资助项目(211057);上海市系统科学一流学科建设项目(XTKX2012);上海市青年科技启明星计划资助项目(A)类(11QA1404500)

摘　　要：为了研究Pearson负相关性信息对协同过滤算法的影响,提出了一种考虑负相关性信息的协同过滤算法。该算法选取正相关用户作为最近邻居,负相关用户作为最远邻居,使用参数调节最近邻居和最远邻居在推荐过程中的作用。MovieLens数据集上的对比实验表明,负相关性不仅可以提高推荐结果的准确性,而且可以增加推荐列表的多样性;进一步分析发现,负相关性还可以大幅度提高不活跃用户的推荐准确性。该工作表明,负相关性有助于解决推荐系统中准确性、多样性两难的问题和冷启动问题。In order to study the effect of the negative correlation of Pearson to collaborative filtering algorithm, this paper presented an improved collaborative filtering algorithm. Firstly this algorithm selected the positive and negative correlation users as the nearest neighbor set and furthest neighbor set respectively, and then made use of a tunable parameter to adjust the effect of the nearest and furthest neighbor set on recommendation. The experiment results on MovieLens dataset show that negative correlations can not only significantly improve the accuracy of recommendations, but also increase the diversity of recommendation lists. It has found that collaborative filtering algorithm by considering the negative correlations can greatly improve the recommendation accuracy of users with small degrees. This work suggests that the negative correlations help solve the dilemma of accuracy and diversity and cold start problem of the recommender systems.

关 键 词：协同过滤 负相关性 最远邻居集 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] TP301.6[自动化与计算机技术—计算机科学与技术]