基于进化博弈理论的关联规则聚类方法研究

Associate Rules Cluster Approach Based on Evolutionary Game Framework Theory

聚类是可视化分析领域压缩数据的有效方法。针对关联规则可视化挖掘中高阶关联特性和无法预先确定分簇数量的问题,设计一种基于进化博弈理论的关联规则聚类方法。首先采用超图工具描述规则间的关联特性;其次,将超图聚类的动态过程建模为进化博弈理论,证明均衡点与优化问题解的一一对应关系,并推导出用于求解聚类的动力学方程。仿真试验表明,该算法能够在未知分簇数量的前提下实现准确聚类,并具有较强的鲁棒性。

Clustering is an effective method to compress data in the visualization area. In order to solve the challenge upon high-order similarities and previously unknown number of classes in visual association rule mining, a cluster approach based on the evolutionary game theory is proposed. Firstly, the hypergraph method is used to describe the high-order similarities among rules. Secondly, the dynamic process of hypergraph clustering is modeled as evolutionary game theory, and the corresponding relation between the equilibrium point and the solution of the optimization problem is proved, and the dynamics for iterative clustering is derived. Finally, simulation results over synthetic data show that the proposed algorithm can achieve accurate hierarchical clustering under the condition of unknown number of clusters, and has strong robustness.