Diffusion over the network using latent variable model

This thesis investigates how the information spreads over the networks with the known structure dependent on the character of the connections. First, the dissimilarity scores between individuals within the networks are calculated. Then the probabilities of a strong contact are modelled using the EM algorithm. Finally, the obtained probabilities are applied as the trans­mission rates to the SI contagion model. Additionally, there are three arti­ficial Erdos-Renyi networks with different sparsity created to show whether the sparsity of links plays an important role in the diffusion process. The approach proposed in the paper does not rely only on the fixed transmis­sion rate independent from the nodes and vertices in the networks. It also tries to incorporate the individual-specific transmission rates, which vary per each connection. The thesis finds that the information speed and range may depend on where it starts. Firstly, the propagation of the information in the artificial networks shows that the sparsity can be an important factor in the range of the diffusion. Moreover, remote individuals can be not good starting points of the diffusion, whereas the initialization in the clique of­fers higher diffusion range on average. Individual-specific transmission rates may also ensure that the diffusion does not arrive to all of the individuals. Additionally, the individual-specific probability offers on average a smaller range of the diffusion.