基于改进LPA算法的去中心化电子商务网络消费者类别划分算法  

A Decentralized Classification Algorithm for Online Consumers Based on Improved LPA

作　　者：刘铸 钱晓东[2] Liu Zhu;Qian Xiaodong(School of Electronic and Information Engineering,Lanzhou Jiaotong University,Lanzhou 730070,China;School of Economics&Management,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学电子与信息工程学院,兰州730070 [2]兰州交通大学经济管理学院,兰州730070

出　　处：《数据分析与知识发现》2023年第10期25-36,共12页Data Analysis and Knowledge Discovery

基　　金：国家自然科学基金项目(项目编号:71461017)的研究成果之一。

摘　　要：【目的】针对LPA算法应用在去中心化电子商务网络消费者类别划分中标签传播距离以及节点相似度判断的不足,提出一种基于改进LPA的类别划分算法。【方法】首先,针对LPA依据共有关系度量节点远近的问题,引入余弦相似度公式度量节点相似度,并构建相似度邻接矩阵,为了符合局部性的特点和减少度量的时间复杂度,引入反查表原理。其次,运用度中心性指标选出初始中心点,并以集聚系数指标为标签更新规则,根据LPA中标签传播特点不符合局部性要求,提出标签传播距离优化公式。【结果】在两种邻居相似度阈值的网络中,改进LPA比传统LPA的类别结构模块度Q值分别提高了0.054和0.145;在不同规模的数据上模块度Q值至多提高了0.092。【局限】需要主观设定两个参数和使用反查表,时间复杂度与网络规模为平方关系。【结论】改进的LPA能更好地限制标签传播,使得类别内节点相似度更高、类别间节点相似度更低,适用于去中心化电子商务消费者网络。[Objective]This paper proposes a classification algorithm based on the improved LPA model,aiming to improve the label propagation distance and node similarity judgment in decentralized e-commerce network consumer classification.[Methods]Firstly,we introduced the cosine similarity formula to measure the similarity of nodes and constructed a similarity adjacency matrix.These steps improved the measurement of node distance according to their shared relationship in the LPA algorithm.We also introduced the principle of a back lookup table to conform to the characteristics of locality and reduce the time complexity.Secondly,we selected the initial center point with the degree centrality index and used the clustering coefficient index to update the label rules.We proposed the label propagation distance optimization formula to make the LPA algorithm meet the locality requirements.[Results]The category structure modularity Q of the improved LPA algorithm was 0.054 and 0.145 higher than the traditional LPA algorithm in the network with two neighbor similarity thresholds.The modular Q value increased up to 0.092 on data of different scales.[Limitations]The paper needs to set two parameters and use the principle of the back lookup table.The relationship between time complexity and network size is square.[Conclusions]The improved LPA can more effectively limit label propagation,which creates higher intra-categories node similarity and lower inter-categories node similarity.It is suitable for analyzing decentralized e-commerce consumer networks.

关 键 词：去中心化 电子商务 消费者网络 局部性 标签传播 类别划分 

分 类 号：TP393[自动化与计算机技术—计算机应用技术] G350[自动化与计算机技术—计算机科学与技术]