Browsing by Author "Kaski, Kimmo K., Prof., Aalto University, Finland"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Modelling Influence Spreading on Complex Networks
    (Aalto University, 2022) Kuikka, Vesa; Kaski, Kimmo K., Prof., Aalto University, Finland; Tietotekniikan laitos; Department of Computer Science; Complex Systems; Perustieteiden korkeakoulu; School of Science; Lampinen, Jouko, Prof., Aalto University, Department of Computer Science, Finland
    Network Science is a growing field of research and it has gained research interest in many application areas. One definition of network science is the study of network representations of physical, biological and social phenomena aiming at predictive models of these phenomena. In addition to specialised methods of network modelling, networks can be studied with general methods of computer science, statistics and many other fields of mathematics.    In this thesis I have developed an influence spreading model that can be applied to social networks, epidemic spreading, and spreading processes in physical and biomedical networks. The model is based on detailed network structures of nodes, links and paths in the network. Individual node and link weights are interpreted as probabilities of transferring information, influence or infection over nodes and via links, respectively.    The model is demonstrated with social networks, epidemic spreading in structured organisations and interoperability in brain networks. In this context, I have proposed a community detection method to discover groups or connected regions in network structures, especially in social networks. In the case of brain networks, I have also discussed how the influence spreading model can be exchanged with a network connectivity model. This demonstrates how the community detection method can be used with different network models. The proposed community detection method and novel metrics are based on the influence spreading model presented in this study.    The community detection method detects overlapping and hierarchical structures. Node and link weights can be used to describe weak and strong interactions or different levels of granularity of results. The method searches the local maxima of an objective function in detecting different splittings of the network. The objective function that I have used in the algorithm for detecting communities and sub-communities can be used as a quality function to compare the cohesion between any sets of nodes in the network structure. In addition, I have proposed alternative quality functions for measuring the probability of formation and robustness of composition for different community structures. Different probabilistic temporal distributions can be implemented in the spreading model. Temporal spreading has been demonstrated with two different distributions, the Poisson distribution and a probabilistic distribution describing temporal delays in message forwarding events. In order to investigate epidemic spreading in structured organisations using the influence spreading model I have used two main versions: complex contagion and simple contagion. The complex contagion model allows loops and breakthrough infection via nodes. The simple contagion model does not allow loops or breakthrough infection. It is assumed that the complex contagion model can describe the spread of virus infections where the main method of spreading is through the air with droplets and particles and breakthrough infection has a large effect.