高级检索
    刘琳岚, 谭镇阳, 舒坚. 基于图神经网络的机会网络节点重要度评估方法[J]. 计算机研究与发展, 2022, 59(4): 834-851. DOI: 10.7544/issn1000-1239.20200673
    引用本文: 刘琳岚, 谭镇阳, 舒坚. 基于图神经网络的机会网络节点重要度评估方法[J]. 计算机研究与发展, 2022, 59(4): 834-851. DOI: 10.7544/issn1000-1239.20200673
    Liu Linlan, Tan Zhenyang, Shu Jian. Node Importance Estimation Method for Opportunistic Network Based on Graph Neural Networks[J]. Journal of Computer Research and Development, 2022, 59(4): 834-851. DOI: 10.7544/issn1000-1239.20200673
    Citation: Liu Linlan, Tan Zhenyang, Shu Jian. Node Importance Estimation Method for Opportunistic Network Based on Graph Neural Networks[J]. Journal of Computer Research and Development, 2022, 59(4): 834-851. DOI: 10.7544/issn1000-1239.20200673

    基于图神经网络的机会网络节点重要度评估方法

    Node Importance Estimation Method for Opportunistic Network Based on Graph Neural Networks

    • 摘要: 机会网络(opportunistic network)是一种利用节点移动的相遇机会实现通信的自组织网络,机会式的通信方式导致其具有时变性与动态性,节点重要度的评估是研究机会网络信息传播的关键.提出一种基于图神经网络的机会网络节点重要度评估方法.将机会网络进行时间切片,对得到的机会网络单元采用聚合图建模,以表征网络信息;采用动态网络嵌入模型提取机会网络单元间的时序变化信息、拓扑结构信息,得到网络的动态属性特征;借助图神经网络(graph neural network, GNN)在图数据处理上的优势,获得网络动态属性特征与节点重要度之间的映射关系,实现节点重要度的评估.在3个真实机会网络数据集MIT,Haggle,Asturias-er上的实验结果表明:相比于时效介数(temporal betweeness, TB)方法、时效度(temporal degree, TD)方法、时效PageRank(temporal PageRank和f-PageRank)方法以及kshell-CN方法,该方法具有更快的消息传播速率和更大的消息覆盖范围,其SIR和NDCG@10指标更优.

       

      Abstract: Opportunistic network is a type of self-organized networks which uses the opportunity of a node moving to realize communication.Because of opportunistic communication mode, opportunistic network has observable time-varying and dynamic characteristics.The estimation of node importance is the key to study the information dissemination of opportunistic network.A novel node importance estimation method based on graph neural network (GNN-NIE) framework is proposed.Opportunistic network is sliced into opportunistic network units which is modeled by aggregate graph to present network information.The dynamic network embedding model is employed to extract the temporal and structural information among the opportunistic network units, so as to obtain the dynamic attribute features of each node in the network.Taking advantage of the GNN’s ability of extracting the features of graph data, the relationship between node dynamic attribute features and the node importance is achieved, so that the node importance of opportunistic network is estimated.The results on three real opportunistic network datasets MIT reality, Haggle project and Asturias-er show that compared with the temporal degree, temporal betweenness, temporal PageRank, and kshell-CN, the proposed method has faster propagation rate, larger message coverage and better SIR and NDCG@10 values.

       

    /

    返回文章
    返回