The Study Of Information Fusion In Identification Of Influent Nodes In Multilayer Networks

Over the past few decades, the research on complex networks made a lot of achievements, a large number of scholars made their effort and energy into academics to allow us to understand more clearly about the composing principle, the mechanism of evolution and other features of complex networks. In the real world, most of the complex systems can be modelled as complex networks, but the research of complex networks has been focused on the characteristics of the networks with single layer. However, the real complex systems often have different kinds of relationships between each unit, they are not independent. On this condition, the single layer networks can’t be used to model this kind of systems. Therefore, a new type of network model, which is called multilayer networks (also called network of networks) has been proposed to describe the complex systems with different kinds of relationship. The multilayer networks is not only an extension of the existing network models, but also an improvement in the network theory.In the previous research on the single layer complex networks, with numbers of explo-ration, researchers found that many global and local information were contained in the nodes of complex networks. In order to make the research more effective, identifying the influential node in complex networks has become very important and significant. Similarly, identifying the influential node in multilayer networks is also theoretical and practical significant. The current funding indicate that the catastrophic risk of multilayer networks is much higher than the risk in independent systems. A seemingly harmless disturbances like ripples may adversely affect the general diffusion resistance. Sometimes this kind of effect may cause a loss of millions or even billions dollars, such as the stock market crash, course closes leaded by volcanic eruptions and so on. Therefore, the evaluation of the influential node in multilayer networks is quite important both on theoretical analysis and practical applications. However, due to the special structure and properties of multilayer networks, identifying the influential node in multilayer networks is more complex than that in single layer networks, because the existing evaluating methods of traditional complex networks can’t be applied in the multi-layer networks. In order to solve this problem, a new evaluating model based on information fusion has been proposed to identify the influential node in multilayer networks.The main work of this paper is summarized as follows:(1) Study the structure characteristics of the multilayer networks and translate the net-works into effective computer language, then the multisource information fusion technology is used to identify the influential node in multilayer networks. In this process, first the multilayer networks should be divided into several single layer networks by different relationship, each single unit will be regarded as one kind of evidence from different sources. Next the Dempster combination rules of Dempster-Shafer evidence theory are used to combine these information from every units of networks to obtain the comprehensive information of the multilayer net-works. The proposed method in this paper can make the evaluation results more accurate and effective by combining the information of different layers of networks.(2) The proposed evaluation model for identifying influential node in the multilayer net-works in this paper is based on the information fusion technology and the D-S evidence theory is used as a tool in this model. So the most critical step in the processing of building the model is how to construct an effective basic probability assignment under this background, which is also an important issue in the research on evidence theory. In order to solve this problem, considering the research background of this paper and the structure of the multilayer net-works, a new method based on the relevance matrix is proposed in this paper to construct the basic probability assignment. The relevance matrix can be obtained from the distance matrix, which can be calculated based on the network information with certain method. The con-struction of the basic probability assignment can fully reflect the corresponding relationship among the nodes in the networks. The proposed method for constructing the basic probability assignment has reference value both in the multilayer network field and the evidence theory field.(3) Verify the proposed model with real data in the daily life. Another two evaluation method existing methods, namely the random walk betweenness centrality and the closeness centrality, are used to compare with the proposed method. The results show that the proposed evaluation model in this paper is effective and it provide a new idea and method of research on the identification of influential node in the multilayer networks. The compared results also show that the more layers of network, the more effective the proposed method becomes.