Research On Node Mining And Network Disintegration Methods Of Epidemic Contact Network

The outbreak of New Coronavirus pneumonia(COVID-19)at the end of 2019 is the most serious global influenza pandemic since the 1918 pandemic.It is also a global public health emergency.To control the outbreak of the epidemic,how to build the epidemic contact network,find the super transmission source in time,and cut off the transmission route of pathogens are the key problems to be solved by the prevention and control department to slow down the spread of the epidemic.On the one hand,according to the28 laws of network communication,the epidemic spread is mainly driven by key node populations and sensitive populations.Therefore,mining and identifying key populations in the epidemic contact network and focusing on prevention and control is of great value to control the epidemic spread.On the other hand,destroying the key nodes of the epidemic contact network and rapidly weakening the transmission capacity of the network,will help to curb the pandemic.Therefore,based on the theoretical methods of key node mining,link prediction,and network disintegration,this paper makes a systematic and in-depth study on the epidemic contact network modeling,key node mining,and network disintegration,to help the prevention and control department accurately identify and deploy.The main work of this paper is as follows:(1)Aiming at the construction of an epidemic contact network,a construction framework of an epidemic contact network is designed based on the large-scale real interpersonal contact data obtained by WiFi sensors.The framework extracts personal trajectory information according to time series and integrates the trajectory information into a contact network that changes dynamically with time.At the same time,considering the situation of environmental infection,the concept of asymmetric contact is proposed.The symmetric contact diagram and asymmetric contact diagram are combined to generate the final epidemic situation contact diagram.(2)Aiming at the key node mining of epidemic contact network,a key node mining algorithm TPGAN(Topological Potential models based on Graph Attention Networks)based on graph neural network and topological potential is designed.TPGAN uses graph neural networks to assign weights to connected edges based on the attribute feature vectors of the nodes,and calculates the topological potential values of each node from the weight values,treating the key nodes as local high potential areas of the topological potential field,and thus obtaining the key nodes in the epidemic contact network.Finally,the SEIR(Susceptible-Exposed-Infectious-Recovered)model is used to simulate the spread of the virus,and adaptive intervention experiments are carried out on the key nodes mined by different node importance measurement methods.The results showed that the total infection rate of the TPGAN method decreased by 10.2% compared with random isolation,which verified the effectiveness of this method.(3)Aiming at the problem of mining key nodes in an epidemic network under incomplete information,a directed network link prediction algorithm LPDN(Link Prediction for Directed Networks based on GAT)based on an attention mechanism is designed.LPDN encodes the nodes into vectors through the encoder,and the decoder generates edge vectors through the aggregation nodes and applies the scoring function to calculate the existence probability of the link between the two nodes.LPDN is used to reconstruct the network to complete the missing links in the network,and then the reconstructed network is used to identify the nodes.The experimental results show that network reconfiguration helps to improve the mining effect of key nodes,which is6.7% higher than that before reconfiguration.(4)Aiming at the intelligent disintegration of the epidemic transmission chain,a directed network intelligent disintegration method DDGD(Directed network Disintegration strategy based on Graph-Embedding via DQN)based on graph embedding and deep reinforcement learning is proposed.In this method,the direction information and structure information of the network is represented in a low dimensional embedding space by graph embedding,and the node embedding vector and graph embedding vector is generated.Combining reinforcement learning and deep neural networks,DDGD is used to automatically learn the strategy of optimization objectives,to more effectively realize the intelligent disintegration of the directed network,and its disintegration effect is improved by 28.7%compared with the baseline method.DDGD is not limited to the epidemic contact network.DDGD has good mobility and provides a scalable solution for directed network disintegration.