Dynamics of information diffusion on online social networks

Abstract

Understanding and modeling diffusion of information on social networks is a widely studied research problem. Diffusion models have evolved from simple epidemiological models to complex models which are based on network structure and voluminous online data. We aim at understanding and studying information diffusion on Online Social Networks(OSNs). Thetwomajoraimsofthethesisare: (i)Bettermodelingofinformation diffusion in OSNs and (ii) Developing efficient algorithms for accelerating virality in OSNs. Wefirstproposeanartificialframeworkformodelingthespreadofmemesonsocial networkingsites(SNSs). Theframeworkconsistsoftwomajorcomponents,(1)asynthetic networkstructurallyresemblingtherealworldsocialnetworkshavingpropertieslikescalefree degree distribution, community structure and core-periphery structure and (2) an information diffusion model based on heterogeneity of edges in the synthetic network. We show that, both, the network and the spreading model, are necessary for simulating a real world meme propagation. The framework has been validated with the help of real world diffusion data of Higgs boson meme on Twitter. Next, we highlight the important differences between the diffusion of a social networking site (SNS) on the web and diffusion of an Internet meme on an SNS. We then propose a model for the shifting behavior of users from an old SNS to a new SNS when the new SNS is introduced on the web. Two major parameters considered in the model are (i) Stability/attachment factor associated with the old SNS and (ii) Feature space of the new SNS. It is shown that it is important for a new SNS to possess at least one new feature to make its mark on the web. The model is validated with the help of real world datasets and a detailed survey. Lastly, based on the observations from the above models, algorithms for accelerating virality on SNSs are proposed. These algorithms, based on a hill climbing approach, direct a meme’s path from periphery to core of the network, where periphery and core are determined with the help of the k-shell decomposition algorithm. Experimental results show that the proposed algorithms are a lot more efficient in comparison to random walk and the well known Adamic walk. The worst case time complexity of the proposed algorithm is linear in terms of the number of nodes in the network. The study also shows the presence of pseudo-cores in the network - the nodes which are equally influential as the core nodes but more susceptible, which makes them perfect candidates to be the seed nodes for the spread of a meme.