摘要
为提高多样化复杂网络中影响力节点识别的准确性和鲁棒性,提出一种基于深度学习的多样化复杂网络影响力节点识别方法。首先,采用多个中心性指标从不同方面评估节点在网络拓扑结构中的重要性,通过可学习权重向量自适应地决定不同复杂网络中各指标的权重;接着,提出一种支持不同特征维度的Transformer框架;最后,利用Transformer模型对不同距离的邻居信息进行分级聚合,以提取邻域的上下文信息。在多种复杂网络数据集上完成了验证实验,结果表明,所提方法在不同规模、不同类型的复杂网络上均取得了较好的影响力节点识别性能,有效提高了影响力节点识别的准确性和鲁棒性。
To improve the accuracy and robustness of influential node recognition in diverse complex networks,a deep learning-based recognition method for influential nodes in diverse complex networks was proposed.Firstly,multiple centrality indexes were utilized to evaluate the importance of network topology from different perspectives,the weight of each index in different complex networks was decided adaptively through the learnable weight vector.Secondly,a new Transformer framework that could handle features of different dimensions was proposed.Finally,the Transformer model was exployed to realize hierarchical aggregation of the neighbor information in different distances,so as to extract the contextual information of the neighborhood.Validation experiments were carried on multiple complex network datasets,the results showed that the proposed method achieved a good recognition performance of influential nodes for the complex networks of different scales and different categories,effectively improving the accuracy and robustness of influential node recognition.
作者
马玉磊
郭莎莎
MA Yulei;GUO Shasha(Department of Computer and Information Engineering,Xinxiang University,Xinxiang 453000,China)
出处
《电信科学》
北大核心
2025年第6期154-165,共12页
Telecommunications Science
基金
河南省科技厅重点研发与推广专项(科技攻关)(No.212102210405)
新乡学院教育教学改革研究与实践项目(No.31)。
关键词
复杂网络
深度学习
自注意力机制
中心性度量
影响力节点
complex network
deep learning
self-attention mechanism
centrality measurement
influential node
作者简介
通信作者:马玉磊(1982-),男,新乡学院计算机与信息工程学院副教授,主要研究方向为计算机网络与网络分析。mayulei8219@163.com;郭莎莎(1993-),女,新乡学院计算机与信息工程学院助教,主要研究方向为图像处理。
