| With the development of society and technology,network data become more and more prevalent.The complex system of the real world,including social relationship,ecosystem and online social platform can be well abstracted by networks.Understanding the nature and laws of these complex and varying networks,and exploring valuable information are new challenges brought by the era of big data.After the development of network science for nearly 20 years,researchers’ focus has gradually turned to networks composed of multiple layers of network.One of the most important issues is how to distinguish which layers are more critical in a multi-layer network.Based on complex network theory,random walk theory,information theory and graph neural network,this paper focuses on extract key network layers playing important role in structure,property and function in a multi-layer network.The main contribution of this paper is be briefly summarized in the following outline.(1)The cooperation and competition among network layers in multi-layer networks are investigated,and then the influence of network layers on the connectivity of the whole network is analyzed quantitatively.Finally,a method to measure the importance of network layer structure is proposed.The experiment results show that our method can more effectively reflect the impact of the layer network layer on the overall connectivity and can mine the information that is neglected by the classical method.(2)The dynamic between network layers in multi-layer network is investigated and a more accurate and robust method to measure the change of mesoscopic structure in network layer is proposed by using the network random walk technique.Finally,this method is used to extract the network layers whose mesoscopic structure have changed significantly.Experiment results in various network models show that our metrics are more sensitive to important meso-structural changes and more robust to network noise.(3)The importance of network layer in multi-layer network classification scenarios is investigated,and a method to quantify the influence of network layers on classification tasks is proposed.In the task of multi-layer network classification,some network layers may contain effective information,while some network layers may not.In this paper,a multi-layer network node classification model based on neural network,which gives different significance to different layers in the sense of classification accuracy,is proposed.At the same time,in order to verify the effectiveness of the method,this paper proposes a general method to estimate the classification information contained in the network structure from the perspective of information theory.The experimental results show that our multi-layer network classification model can effectively extract key network layers while achieving the state-of-the-art classification result. |
