基于二分网络投影的多维度推荐算法设计研究  

MULTI-DIMENSIONAL RECOMMENDATION ALGORITHM BASED ON BIPARTITE NETWORK PROJECTION

作　　者：熊湘云[1] 伏玉琛[1] 刘兆庆[1] 

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006

出　　处：《计算机应用与软件》2014年第8期253-256,共4页Computer Applications and Software

基　　金：国家自然科学基金项目(61070122)

摘　　要：二分网络投影推荐算法明显优于传统的协同过滤推荐算法。为综合考虑二分网络顶点的相关信息,以解决数据稀疏性引起的推荐不准确问题,并提高推荐结果的多样性,提出多维度推荐算法BNPM(Bipartite-Network-Projection-based multi-dimensional recommendation algorithm):综合基于项目、用户和属性投影三个维度的推荐模型,将项目的类别信息以向量的形式由项目顶点投影至用户顶点构建基于属性投影的推荐模型,并根据推荐算法的评价标准,动态自适应地确定三个维度推荐模型的权因数值产生最终推荐。实验结果表明,BNPM推荐算法可以有效提高推荐结果的准确性和个性化程度。Bipartite network projection recommendation algorithm is distinctly superior to traditional collaborative filtering recommendation algorithm. In order to consider the relevant information of bipartite network vertexes comprehensively,so as to solve the problem of inaccurate recommendation caused by data sparse and to improve the diversity of recommendation results,we propose in this paper a multi-dimensional recommendation algorithm named BNPM（bipartite network projection-based multi-dimensional recommendation algorithm）. It integrates the recommendation models in three dimensions： the item-based projection,the user-based projection and the attribute-based projection,projects the category information of items from items vertexes to users vertexes in form of vector to construct the attribute projection-based recommendation model. And according to evaluation standard of the recommendation algorithm,it dynamically and adaptively determines the weight factor value of the three dimensions recommendation algorithm to generate final recommendation. Experimental results show that BNPM recommendation algorithm can effectively improve the accuracy and personalisation degree of recommendation results.

关 键 词：二分网络 投影 多维度 推荐技术 

分 类 号：TP309.7[自动化与计算机技术—计算机系统结构]