多智能体系构架下的属性图分布式聚类算法

发布时间：2018-01-26 23:00

  本文关键词： 属性图聚类 集群形成博弈 紧密性和均匀性约束 分布式学习算法 多智能体系统　出处：《计算机科学》2017年S1期 　论文类型：期刊论文

【摘要】：近年来属性图聚类受到了广泛关注,其目的是将属性图中的节点划分到若干簇中,使得每一个集群都有紧密的簇内结构和均匀的属性值。现有的理论主要是假设属性图中的节点或对象是为了协助优化某个给定的方程,而忽略了它们在现实生活中本身的属性。同时,一些开放性问题尚未得到有效解决,如异构信息集成、计算成本高等。为此,把属性图聚类问题理解为自身节点代理的集群形成博弈。为了有效地整合拓扑结构和属性信息,提出了基于紧密性和均匀性约束的节点代理策略选择。进一步证明了博弈过程将会收敛到弱帕累托纳什均衡。在实证方面,设计了一个分布式和异构的多智能体系统,给出了一个快速的分布式学习算法。该算法的主要特点是结果分区的重叠率可以由一个事先给定的阈值控制。最后,在现实社交网络上进行了模拟实验,并与目前先进方法进行比较,结果证实了所提算法的有效性。
[Abstract]:In recent years, attribute graph clustering has received extensive attention, the purpose of which is to divide the nodes in the attribute map into a number of clusters. So that each cluster has a tight cluster structure and uniform attribute value. The existing theory mainly assumes that the nodes or objects in the attribute map are to help optimize a given equation. At the same time, some open problems have not been solved effectively, such as heterogeneous information integration, high computing cost and so on. In order to integrate topology and attribute information effectively, the clustering problem of attribute graph is understood as the cluster game of its own node agent. The selection of node agent strategy based on compactness and uniformity constraints is proposed. It is further proved that the game process will converge to the weak Pareto Nash equilibrium. A distributed and heterogeneous multi-agent system is designed and a fast distributed learning algorithm is proposed. The main feature of the algorithm is that the overlap rate of the result partition can be controlled by a predetermined threshold. Finally. The simulation experiments on real social networks are carried out and compared with the current advanced methods. The results show that the proposed algorithm is effective.
【作者单位】： 中央财经大学管理科学与工程学院;
【基金】：国家自然科学基金项目(71401194,71401188) 中央财经大学“青年英才”培育支持项目(QYP1603)资助
【分类号】：TP18;TP311.13
【正文快照】： 本文受国家自然科学基金项目(71401194,71401188),中央财经大学“青年英才”培育支持项目(QYP1603)资助。1引言许多现实中的信息系统是由大量高度关联的参与者或对象组成的,如在线社会网络、无线传感器网络和众包平台。这些系统可以被属性图很好地模拟出来,其中节点代表组件对，

本文编号：1466839