面向用户偏好分析的无向图层次聚类并行优化算法  

A Parallel Optimization Algorithm for Hierarchical Clustering of Undirected Graph Oriented to User Preference Analysis

作　　者：刘晓慧 江峰[1] 杜军威[1] 余东瑾 LIU Xiaohui;JIANG Feng;DU Junwei;YU Dongjin(College of Information Science and Technology,Qingdao University of Science and Technology,Qingdao 266061)

机构地区：[1]青岛科技大学信息科学技术学院,青岛266061

出　　处：《计算机与数字工程》2020年第5期1137-1142,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61273180,61402246);山东省自然科学基金项目(编号:ZR2018MF007)资助。

摘　　要：用户偏好分析是构建用户个性化服务的基础,是用户在对商品或者服务进行考量之后所做出的理性且带有个人倾向的选择。传统的无向图层次聚类算法在聚合时每次都需要重新运算边的权重,而且每次合并的点数量有限。上述问题使得传统算法不能被有效用于用户偏好分析。为能精确地挖掘用户偏好,文中提出一种无向图层次聚类的并行优化算法。首先,算法通过分裂高热节点,削弱衰减因子的消极影响;其次,采用一种并行的方法对无向图实现聚合,优化聚类的速度;最后,将基于用户的搜索行为而得到内容偏好分类判断为依据对内容进行聚合。通过实验对该算法进行了验证,实验结果表明该算法能够显著提高聚类的覆盖率、效率及准确率。User preference analysis is the basis for constructing user personalized services.It is a rational and inclined choice made by users after considering products or services.The traditional hierarchical clustering algorithms of undirected graph need to recalculate the weights of edges each time during the aggregation,and the number of points merged is limited.Because of the above problems,the traditional algorithms can not be applied to user preference analysis effectively.In order to accurately mine user preferences,a parallel optimization algorithm for hierarchical clustering of undirected graphs is proposed.First,the algorithm weakens the negative effects of the attenuation factor by splitting the high-heat nodes.Secondly,a parallel method is used to aggregate the undirected graphs to optimize the speed of clustering.Finally,The content preference classification is determined based on the user’s search behavior as a basis for aggregating the content.The algorithm is verified by experiments,and the experimental results show that the algorithm can obviously improve the coverage,efficiency and accuracy of clustering.

关 键 词：用户偏好 无向图层次聚类 并行优化 聚合 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]