基于改进DeepWalk算法的电商网络链路预测研究  

Research on E-Commerce Network Link Prediction Based on Improved DeepWalk Algorithm

作　　者：钱晓东[1] 史玉林 郭颖 Qian Xiaodong;Shi Yulin;Guo Ying(School of Economics and Management,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]兰州交通大学经济与管理学院,兰州730070

出　　处：《数据分析与知识发现》2024年第12期62-72,共11页Data Analysis and Knowledge Discovery

基　　金：甘肃省自然科学基金项目(项目编号:23JRRA898)的研究成果之一

摘　　要：【目的】针对DeepWalk算法应用在电子商务网络链路预测中的不足,提出一种基于改进的DeepWalk算法的链路预测算法。【方法】针对DeepWalk算法随机游走过程中平等对待每个节点的问题,利用电商网络的结构和属性信息对随机行走进行偏置,引导游走过程更有针对性地遍历图中不同类型的节点;传统的DeepWalk算法使用余弦相似度方法进行节点的相似度度量,该方法不能很好地表现用户和商品关系的问题,本文将巴氏(Bhattacharyya)系数引入现有的非线性相似度计算中,创建新的节点相似度方法。【结果】改进后的DeepWalk算法在不同规模的数据中平均召回准确率提高范围在0.05~0.17;在计算节点相似性时,节点属性相似度贡献α的最优值在0.5~0.6之间。【局限】随着时间复杂度的增长可能会导致算法可扩展性下降。【结论】经过改进的算法能够更好地学习节点嵌入向量,以识别电商网络中的节点相似性,从而显著提升节点表示的准确性和链路预测的效果。[Objective]This paper aims to address the deficiency of DeepWalk algorithm in link prediction of ecommerce network,a link prediction algorithm based on improved DeepWalk algorithm is proposed.[Methods]According to the problem that the traditional DeepWalk algorithm treats each node equally in the random walk process,the structure and attribute information of the e-commerce network are biased to the random walk,so as to guide the walking process to traverse different types of nodes in the graph more targeted.This paper solves the problem that the traditional DeepWalk algorithm can not well represent the relationship between users and commodities by using cosine similarity measurement method,Bhattacharyya Coefficient is introduced into the existing nonlinear similarity calculation model to create a new similarity model.[Results]Experimental results show that the average recall accuracy of the improved DeepWalk algorithm is improved by a maximum of 0.17 and a minimum of 0.05 in different data scales.When calculating the node similarity,the optimal value of the node attribute similarity contribution alpha is between 0.5 and 0.6.[Limitations]In this paper,sensitivity parameters must be set subjectively,and the time complexity of the random walk is proportional to the network size N,so the scalability of the algorithm may decrease as the time complexity increases.[Conclusions]It shows that the improved algorithm can learn the node embedding vector well,so as to understand the similarity of nodes in the e-commerce network.

关 键 词：DeepWalk算法 电商网络 链路预测 推荐 

分 类 号：O157.5[理学—数学] F713.36[理学—基础数学]