3. Controlling the Development of Epidemic Diseases Using Trajectory Networks and Social Networks
Project Description: In recent years, COVID-19 has severely affected all aspects of human life. In order to slow down the spread of the virus, governments have introduced various policies. In the context of infectious diseases, it is crucial for policymakers to determine which policies can more effectively mitigate the spread of the epidemic. We consider an infectious disease network, which is a dynamic graph of the causal relationship of the epidemic that changes over time, as shown in the left figure 3. Among them, Z represents time-varying confounding factors, which are factors that affect treatment and diagnosis results; X represents the covariate characteristics of a region in a time period t (such as March) (such as the popularity of residents searching for epidemic-related keywords in search engines); P indicates whether a certain policy is implemented in the region within time period t (1 for yes, 0 for no); Y records the confirmed result; W represents the adjacency matrix of the network distance between regions in time period t. It is assumed that the confounding factor Z has a causal relationship with the policy implementation P and the confirmed result Y at the current time period t, and the policy implementation P, the confirmed result Y, and the confounding factor Z at the previous time period t-1 will also affect the current confounding factor Z. Recurrent Neural Networks (RNNs) can be used to extract useful historical information from data from previous time periods. Then use Graph Convolutional Network (GCN) to extract features of the network structure between regions, mine out stubborn nodes with severe epidemics, and then learn the feature vector of the current confounding factor Z. The key point is that using only the distance between regions as the basic information to form the W matrix is a bit simple and will lose a lot of details. It can be considered to replace the original W matrix with a more informative regional network relationship matrix containing stubborn nodes, as shown in the right figure 3. For some areas with very serious confirmed cases or the source of the epidemic, we believe that they are stubborn nodes in the network relationship, and the weight of the adjacency matrix can reflect the degree of influence of stubborn nodes in the region on the epidemic of other nodes.