楊洋 王俊峰



準(zhǔn)確識別出網(wǎng)絡(luò)中的關(guān)鍵節(jié)點(diǎn)是復(fù)雜網(wǎng)絡(luò)研究的重要內(nèi)容之一.現(xiàn)存的關(guān)鍵節(jié)點(diǎn)識別方法多數(shù)是基于網(wǎng)絡(luò)結(jié)構(gòu)提出的中心性度量方法,識別準(zhǔn)確率低且適用范圍具有局限性.因此本文提出了基于圖卷積網(wǎng)絡(luò)的關(guān)鍵節(jié)點(diǎn)識別方法,不僅考慮了節(jié)點(diǎn)屬性,還考慮了網(wǎng)絡(luò)結(jié)構(gòu)和鄰居節(jié)點(diǎn)結(jié)構(gòu).首先,根據(jù)網(wǎng)絡(luò)圖例數(shù)據(jù)提取多維度特征并構(gòu)建特征向量;其次,將節(jié)點(diǎn)特征向量輸入到GCN層學(xué)習(xí);最后,通過回歸損失函數(shù)計算出最小損失,識別出關(guān)鍵節(jié)點(diǎn).本文選取傳播動力學(xué)中的SIR模擬實(shí)驗(yàn)和牽制控制實(shí)驗(yàn)作為評價方式,在不同類型的真實(shí)網(wǎng)絡(luò)上進(jìn)行驗(yàn)證.結(jié)果表明本文提出的方法在適用范圍和準(zhǔn)確率方面較其他方法更具優(yōu)勢.
關(guān)鍵節(jié)點(diǎn); 復(fù)雜網(wǎng)絡(luò); 圖卷積網(wǎng)絡(luò)
TP301.6A2023.032002
收稿日期: 2022-06-28
基金項(xiàng)目: 基礎(chǔ)加強(qiáng)計劃重點(diǎn)項(xiàng)目(2019-JCJQ-ZD-113); 國家自然科學(xué)基金(U2133208); 四川省青年科技創(chuàng)新研究團(tuán)隊(duì)(2022JDTD0014)
作者簡介: 楊洋(1998-), 河南平頂山人, 碩士研究生, 研究方向?yàn)榫W(wǎng)絡(luò)空間安全. E-mail: 305004556@qq.com
通訊作者: 王俊峰. E-mail: wangjf@scu.edu.cn
Research on key node identification of complex network based on GCN
YANG Yang, WANG Jun-Feng
(College of Computer Science, Sichuan University, Chengdu 610065, China)
Accurately identifying the key nodes in the network is one of the important research topics in complex networks. Most of the existing key node identification methods are based on the centrality measurement method by the network structure, which has low identification accuracy and limited scope of application. A key node identification method, based on Graph Convolutional Network (GCN), is proposed in this paper, which considers not only the node attributes, but also the network structure and neighbor node structure. Multidimensional features are extracted first from the network legend data to construct feature vectors and then the node feature vector is input to the GCN layer for learning. Finally, the minimum loss is calculated with the regression loss function, and the key nodes are identified. In this paper, SIR (Susceptible Infected Removed) is choosed as the evaluation method in the propagation dynamics simulation experiment and Pinning Control experiment, the proposed method is verified on different types of real networks, the results show that the GCN-based method proposed in this paper outperforms other methods in terms of scope of application and accuracy.
Key node; Complex network; Graph convolutional network
1 引 言
在網(wǎng)絡(luò)理論的研究中,生物網(wǎng)絡(luò)、電力網(wǎng)絡(luò)以及通訊網(wǎng)絡(luò)等都被證實(shí)為復(fù)雜網(wǎng)絡(luò)[1].關(guān)鍵節(jié)點(diǎn)[2]是能高度影響復(fù)雜網(wǎng)絡(luò)功能的少數(shù)特殊節(jié)點(diǎn).定位關(guān)鍵節(jié)點(diǎn)對網(wǎng)絡(luò)信息傳遞、網(wǎng)絡(luò)同步、網(wǎng)絡(luò)控制起著至關(guān)重要的作用[3].例如:社交網(wǎng)絡(luò)中權(quán)威賬號對輿論的引導(dǎo)作用明顯;社會網(wǎng)絡(luò)中控制流行病的爆發(fā)點(diǎn)能抑制傳染病大規(guī)模傳播;交通網(wǎng)絡(luò)中挖掘關(guān)鍵樞紐能夠?yàn)橐?guī)劃航……