Dynamical Clustering in Electronic Commerce Systems via Optimization and Leadership Expansion

作     者：Li, Hui-Jia Bu, Zhan Wang, Zhen Cao, Jie 

作者机构：Beijing Univ Posts & Telecommun Sch Sci Beijing 100876 Peoples R China Nanjing Univ Finance & Econ Jiangsu Prov Key Lab E Business Nanjing 210003 Peoples R China Northwestern Polytech Univ Ctr Opt IMagery Anal & Learning Xian 710072 Peoples R China 

出 版 物：《IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS》 (IEEE产业信息学汇刊)

年 卷 期：2020年第16卷第8期

页      面：5327-5334页

核心收录：

学科分类：1201[管理学-管理科学与工程(可授管理学、工学学位)] 0808[工学-电气工程] 08[工学] 0811[工学-控制科学与工程] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Key Research and Development Program of China [2019YFB1405000] Key Area R&D Program of Guangdong Province [2019B010137004] National Natural Science Foundation of China [71871233, 71871109, 91646204, U1803263] Beijing Natural Science Foundation Fundamental Research Funds for the Central Universities [3102019PJ006] Shannxi Natural Science Foundation-Key Program [2019ZDLGY17-07, TII-19-3294] 

主　　题：Computational complexity clustering algorithm dynamical systems electronic commerce systems game theory 

摘      要：In many electronic commerce systems, detecting significant clusters is of great value to the analysis, design, and optimization of the commerce behaviors. In this article, we propose a new dynamical approach to detect the cluster configuration fast and accurately which can be applied to electronic commerce systems. First, we analyze the two-stage game in which the leader group members make contributions prior to the follower group, and propose an exact index, i.e., the leadership, to characterize the key leaders. Then an efficient dynamical system is used to guarantee the cluster configuration converges to an optimal state, which assigns each node to the corresponding cluster based on quality optimization, repeatedly. Our method is of high efficiency-the exponential term in the proposed dynamical system makes the convergence to be very fast with a nearly linear time. Extensive experiments on multiple types of datesets demonstrate the state-of-the-art performance of proposed method.