一种融合引力影响的新的矩阵分解推荐方法  

A Novel Appocach to Matrix Factorization Recommender System Using Gravitational Impacts

作　　者：李鹏澎 肖如良[1] 邓新国[2] 林丽玉[1] 蔡声镇[1] 

机构地区：[1]福建师范大学软件学院,福州350117 [2]福州大学数学与计算机学院,福州350116

出　　处：《小型微型计算机系统》2015年第4期696-700,共5页Journal of Chinese Computer Systems

基　　金：教育部规划基金项目(11YJA860028)资助;福建省自然科学基金项目(2013J01219)资助

摘　　要：矩阵分解运用于推荐系统受到越来越多的关注,改进算法也层出不穷,这些算法都面临预测精度不足,收敛速度过慢和不适用于大数据量的推荐系统等问题.针对上述问题,提出一种基于引力影响的矩阵分解推荐方法,物品影响力对推荐系统提高推荐精度具有重要的意义.基于物品流行度和物品质量的高低对用户评分的影响,通过引入物品对用户的引力,实现基于矩阵分解的引力推荐算法.实验表明:在用户没有任何历史行为的情况下也可以做出比较合理的推荐,迭代20次就能达到较好的收敛效果,且在推荐精度上有了明显的提高,并且该算法可适用于数据量大的推荐系统.Matrix factorization has been paid more and more attention on used in the recommender system,the improved algorithm are proposed endlessly, these algorithms are all confronted with prediction accuracy not enough, convergence too slowly and not applicable to large scale of data recommender system. A Novel appocach to matrix factorization Recommender System using Gravitational im- pacts is proposed to solve these problem, Item influence on the recommender system to improve the recommendation accuracy is of great significance. Item＇s Popularity and item＇s quality affecting the users＇ rating ,by introducing items＇ gravitational impacts the users to propose recommender algorithm based on matrix faetorization. Experimental results show that：even though the user does not have any history of behavior can also make reasonable recommendations ,20 iterations can achieve better convergence effect,and has signifi- cantly improved the accuracy of the recommendations, applicable to large scale of data recommender system.

关 键 词：推荐精度 引力 矩阵分解 流行度 推荐系统 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]