Identifying critical nodes in complex networks via graph convolutional networks
Introduction
Nowadays, many real-world systems can be described by complex networks, the nodes represent the elements and edges stand for the relationships among different elements. It is well-known that many mechanisms such as cascading, spreading, and synchronizing are highly affected by a few critical nodes [1]. Therefore, identifying critical nodes in complex networks has caused great concern in the research community [2], [3], [4], [5]. And the spreading ability of nodes has been applied to solve many problems, such as controlling propagation of information and rumors in social networks, ordering reputation of scientists [6] and identifying leaders in social networks [7].
Identifying critical nodes based on the structure and function of networks is a hot research topic on static networks and plays vital role in controlling the spread of epidemic or negative information [8], [9]. Up to now, many methods have been proposed to measure the importance of nodes in static networks. Among of these, degree centrality [10], k-shell [11], H-index [12], [13] and semi-local centrality [14] select influential nodes by node’s own influences. Based on paths in networks, eccentricity centralities [15], betweenness [16] and closeness [17] have been proposed. And HITs [18], LeaderRank [19] and PageRank [20] are based on eigenvector. Zhang et al. have presented VoteRank [21] to identify a set of spreaders with the best spreading ability. Nodes vote in their neighbors in each turn, and the voting ability of neighbors of elected spreader will be decreased permanently. Li et al. proposed a classified neighbors algorithm [22] by assigning different weights for each class of neighbors and summing up the contributions of neighbors to quantify the nodal spreading capability and further to differentiate the influence of various nodes. Fei et al. [23] thought the mutual attraction between different nodes has been defined in complex network, which is inversely proportional to the square of the distance between two nodes. And proposed a method based on the inverse-square law to identify influential nodes. In order to adjust the decreasing factor adaptively according to the degree of global network, Guo et al. [24] proposed a new method named EnRenew based on node’s entropy to select a set of influential nodes. Lv et al. [25]. proposed an average shortest path centrality to identify critical nodes, in which the relative change of the average shortest path of the whole network is taken into account.
Although there are many methods to measure the influence of nodes in complex networks, most of them can be regarded as a mission to find what kind of structure will make the node more influential. With the development of machine learning, some researchers have proposed the concept of GCNs [26], [27], which aims at learning low-dimensional latent representation of graphs. These representations can be used as features for a wide range of tasks on graphs such as classification, clustering, link prediction and visualization. Inspired by this concept of GCNs, we convert the critical node identification problem in complex networks into a regression problem. Many traditional influential nodes mining algorithms usually directly use the adjacency matrix, which may contain both the intrinsic structure and the redundant information. In order to reduce the noise or redundant information and preserve the intrinsic structure information. By using adjacency matrices of networks and GCNs, we propose a simply yet effectively iterative method named to rank nodes. The performance of the proposed method is compared with degree centrality, betweenness centrality, -shell and H-index in SIR model [28], [29] on nine synthetic and fifteen real networks. The results show that the proposed method in this paper can effectively identify critical nodes which have greater impacts on information spreading in complex networks.
The rest of paper is organized as follows: The related works are presented in Section 2 and the background is presented in Section 3. The proposed method to identify critical nodes based on graph convolutional networks is defined in Section 4. Experiment results and the time complexity are analyzed and discussed in Section 5. Conclusions and future interested research topics are given in Section 6.
Section snippets
Related works
In recent years, with the wide application of machine learning, GCNs have been used in community detection [30], [31], link prediction [32], [33] and etiology of diseases [34]. GCNs are semi-supervised methods for graphs and usually can be trained with task-specific loss through back-propagation like standard CNNs. However, due to the lack of grid structures, standard convolution of image data or text data cannot be directly applied to graphics. So the difficulty of GCNs lies in how to extract
Background
In this section, some background information about networks, convolutional neural networks and graph convolutional networks is given.
Method
We aim to bring convolutional neural networks to bear on the problem of identifying critical nodes in complex networks. We propose a framework for learning representations for arbitrary networks. Similar to convolutional neural network for images, we construct locally connected neighborhoods from the input networks. These neighborhoods serve as the receptive fields of a convolutional architecture, allowing the framework to learn effective network representations. The details of RCNN are
Experiments and discussions
The performance of the is evaluated by SIR spreading model, and compared with well-known existing metrics such as degree centrality, betweenness centrality and -shell in nine synthetic and fifteen real networks.
Conclusion
How to identify critical nodes in complex networks is an interesting and important topic in many applications. Most of previous researches concentrate on finding what kind of structure will make the node more influential. Inspired by the concept of GCNs, by using adjacency matrices of networks and CNNs, is proposed in this paper. According to the experimental results on nine synthetic and fifteen real networks, performs much better than other four benchmark methods in identifying the
CRediT authorship contribution statement
En-Yu Yu: Investigation, Data curation, Methodology, Validation, Visualization, Writing - original draft. Yue-Ping Wang: Writing - review & editing. Yan Fu: Writing - review & editing. Duan-Bing Chen: Supervision, Funding acquisition, Writing - review & editing. Mei Xie: Writing - review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work is jointly supported by the National Natural Science Foundation of China with Grant Nos. 61673085 and 61433014, Science Strength Promotion Programme of UESTC, China (Grant No. Y03111023901014006) and National Key R&D Program of China (Grant No. 2017YFC1601005).
References (67)
- et al.
Identifying influential nodes in complex networks with community structure
Knowl.-Based Syst.
(2013) - et al.
Identifying all-around nodes for spreading dynamics in complex networks
Physica A
(2012) - et al.
Ranking spreaders by decomposing complex networks
Phys. Lett. A
(2013) - et al.
Ranking the spreading influence in complex networks
Physica A
(2013) - et al.
A discount strategy in word-of-mouth marketing
Commun. Nonlinear Sci. Numer. Simul.
(2019) - et al.
Fast influencers in complex networks
Commun. Nonlinear Sci. Numer. Simul.
(2019) - et al.
Identifying influential nodes in complex networks
Physica A
(2012) Centrality in social networks conceptual clarification
Soc. Netw.
(1978)- et al.
The anatomy of a large-scale hypertextual web search engine
Comput. Netw. ISDN Syst.
(1998) - et al.
Identification of influential spreaders based on classified neighbors in real-world complex networks
Appl. Math. Comput.
(2018)
Identifying influential nodes in complex networks based on the inverse-square law
Physica A
A novel measure of identifying influential nodes in complex networks
Physica A
Community detection in multi-layer networks using joint nonnegative matrix factorization
IEEE Trans. Knowl. Data Eng.
Graph regularized nonnegative matrix factorization for temporal link prediction in dynamic networks
Physica A
Extracting stage-specific and dynamic modules through analyzing multiple networks associated with cancer progression
IEEE/ACM Trans. Comput. Biol. Bioinform.
Exploring social structure using dynamic three-dimensional color images
Soc. Netw.
A human protein-protein interaction network: A resource for annotating the proteome
Cell
Cortical hubs form a module for multisensory integration on top of the hierarchy of cortical networks
Front. Neuroinformatics
Quantifying the influence of scientists and their publications: distinguishing between prestige and popularity
New J. Phys.
The small world yields the most effective information spreading
New J. Phys.
Factoring and weighting approaches to status scores and clique identification
J. Math. Sociol.
Identification of influential spreaders in complex networks
Nat. Phys.
The H-index of a network node and its relation to degree and coreness
Nature Commun.
Topological structure and the h index in complex networks
Phys. Rev. E
Eccentricity and centrality in networks
Soc. Netw.
A set of measures of centrality based on betweenness
Sociometry
Authoritative sources in a hyperlinked environment
J. ACM
Leaders in social networks, the delicious case
PLoS One
Identifying a set of influential spreaders in complex networks
Sci. Rep.
Influential nodes identification in complex networks via information entropy
Entropy
A comprehensive survey on graph neural networks
Graph neural networks: A review of methods and applications
Blocking links to minimize contamination spread in a social network
ACM Trans. Knowl. Discov. Data
Cited by (90)
HCCKshell: A heterogeneous cross-comparison improved Kshell algorithm for Influence Maximization
2024, Information Processing and ManagementGraph neural networks-based preference learning method for object ranking
2024, International Journal of Approximate ReasoningIdentifying critical nodes of cyber–physical power systems based on improved adaptive differential evolution
2024, Electric Power Systems ResearchIdentifying influential nodes through an improved k-shell iteration factor model
2024, Expert Systems with ApplicationsOlapGN: A multi-layered graph convolution network-based model for locating influential nodes in graph networks
2024, Knowledge-Based Systems