基于聚类系数的推荐算法  

Modified recommendation algorithm based on clustering coefficient

作　　者：许鹏远[1] 党延忠[1] 

机构地区：[1]大连理工大学系统工程研究所,辽宁大连116024

出　　处：《计算机应用研究》2016年第3期654-656,660,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(71031002)

摘　　要：针对于标准二分图网络推荐算法(NBI)的物质扩散机制过于简单的问题,提出了基于聚类系数的改进NBI算法(简称NBICC)。推荐系统可以被抽象为一个有向加权二分图网络,在物质扩散的过程中,考虑到聚类系数因素的影响,重新定义了商品之间的相似度的计算公式,进而获得了更加精确的推荐结果。Ranking score、precison、recall评价指标被应用在提出的新算法中,实验结果表明,在这三样重要指标上,NBICC算法都强于标准NBI算法。Accordance with the problem that the mass diffusion mechanism which standard NBI algorithm used was too simple,this paper proposed a modified NBI algorithm based on clustering coefficient（ NBICC）. This algorithm regarded recommendation system as a direct graph with weight. In order to obtain more accurate result,it redefined the calculation formula of similarity by considering clustering coefficient in the process of mass diffusion. Numerical results indicate that the algorithmic accuracy measured by the average ranking score,precision and recall is improved greatly.

关 键 词：推荐系统 有向加权图 聚类系数 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP301.6[自动化与计算机技术—控制科学与工程]