基于随机游走和多样性图排序的个性化服务推荐方法  

Personalized Service Recommendation Method Based on Random Walk and Diversified Graph Ranking

作　　者：方晨[1,2] 张恒巍 王娜[1] 王晋东 FANG Chen;ZHANG Heng-wei;WANG Na;WANG Jin-dong(Information Engineering University,Zhengzhou,Henan 450001,China;State Key Laboratory of Mathematical Engineering and Advanced Computing,Zhengzhou,Henan 450001,China)

机构地区：[1]信息工程大学,河南郑州450001 [2]数字工程与先进计算国家重点实验室,河南郑州450001

出　　处：《电子学报》2018年第11期2773-2780,共8页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61309013;No.61303074);河南省科技攻关计划项目(No.12210231003)

摘　　要：针对传统服务推荐算法由于数据稀疏性而导致推荐准确性不高,以及推荐结果缺乏多样性等缺陷,提出基于随机游走和多样性图排序的个性化服务推荐方法(PRWDR).在分析直接相似关系稀疏性的基础上提出带权重的随机游走模型,通过在用户网络上进行随机游走来挖掘更多的相似关系;基于所有相似用户预测服务的Qo S值,并给出服务图模型构建方法,以过滤大量性能过低的候选服务;提出最优节点集合选取策略,利用贪婪算法得到兼具推荐准确性和功能多样性的服务推荐列表.在公开发布的数据集上进行实验,并与多个经典算法进行比较,验证了本算法的有效性.In view of the low recommendation accuracy due to the sparseness of data,and the lack of diversity in traditional service recommendation algorithms,personalized service recommendation method based on random walking and diversified graph ranking(PRWDR)is proposed.On the basis of analyzing the sparseness of direct similarity relationships,a weighted random walk model is proposed,which can excavate more similarity relationships by random walk on the user network.The QoS value of services is predicted based on all similar users,and then the service graph model construction method is presented to filter those services with low performance.By using the greedy algorithm,the optimal node collection selection strategy is proposed to obtain the service recommendation list with both accuracy and diversity.By testing the algorithm on the public dataset and also comparing with several classic algorithms,the validity of PRWDR is verified.

关 键 词：服务推荐 数据稀疏性 多样性 随机游走模型 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]