卷积神经网络的top-k相似节点搜索方法  

Top-k Similarity Node Search Method in Convolutional Neural Network

作　　者：孟祥福[1] 温晶 李子函 纪鸿樟 MENG Xiang-fu;WEN Jing;LI Zi-han;JI Hong-zhang(School of Electronic and Information Engineering,Liaoning Technical University,Huludao 125105,China)

机构地区：[1]辽宁工程技术大学电子与信息工程学院,辽宁葫芦岛125105

出　　处：《小型微型计算机系统》2023年第11期2516-2521,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金面上项目(61772249)资助。

摘　　要：针对复杂网络环境下搜索与目标节点文本和结构均相似的top-k节点问题,本文提出了一种基于卷积神经网络的top-k相似节点搜索算法LRE-CNN.对于一个无向带权复杂网络,首先为网络中每个节点构造基于度和权重的最近邻网络模型,利用最近邻网络相对加权熵计算度和权重对节点结构的影响.然后,通过KL散度比较节点对的差异生成节点结构相似度,从而筛选出目标节点的候选相似节点.最后,利用卷积神经网络(CNN)抽取目标节点和候选相似节点的文本特征间的潜在关系,从而预测出与目标节点文本结构均相似的top-k节点.通过在不同规模的复杂网络上进行实验,并与现有主流相似节点搜索方法进行对比,实验结果表明所提方法具有较高的检索准确率,同时具有较高的执行效率,能够有效适用于大规模复杂网络环境下的相似节点top-k搜索.In order to search top-k nodes with similar text and structure as target nodes in complex network environment,a similar node search algorithm(LRE-CNN)based on convolutional neural network is proposed.For an undirected weighted complex network,a nearest neighbor network model based on degree and weight is constructed for each node in the network,and the relative weighted entropy of nearest neighbor network is used to calculate the influence of degree and weight on node structure.Then,the similarity of node structure is generated by comparing the difference of node pairs through KL divergence,so as to screen out the candidate similarity nodes with the most similarity to the target node.Lastly,convolutional neural network(CNN)is used to extract the potential relationship between the text features of the target node and the candidate similar nodes,and consequentaly the top-k nodes with similar text structure to the target node are predicted.Through experiments on complex networks of different scales and comparison with the existing state-of-the-art similarity node search methods,the experimental results show that the proposed method has high retrieval accuracy and high execution efficiency,and can be effectively applied to the top-k search of similar nodes in large-scale complex network environment.

关 键 词：节点相似度 top-k搜索 卷积神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]