Study On Diffusion Pattern In Complex Social Network

Recently, research on the complex networks has grown rapidly. Complex networks widely existin nature and human society, like food-chain networks, Internet, neural networks and WWW, whichusually have the following properties: power-law degree distribution, small average path length,large cluster coefficient, etc. Now, more and more researches focus on the relationship betweentopological structure and dynamics happened on complex social networks. Specifically, our paperdeeply investigates the following two key problems in complex social networks: the heuristicscheme of selecting influential seeds for maximizing behavior diffusion, and theoretical analysis ofbehavior diffusion pattern based on mean-field theory.The influence maximization problem can be described as follows: if we can try to convince asubset of individuals (so-called initial seeds) to adopt a new behavior and the goal is to trigger alarge cascade of further adoptions, which set of individuals should we target in order to achieve amaximized influence? In this thesis, we propose Pagerank-like heuristic scheme, in whichdiscounting the influence power is adopted to alleviate the “overlapping effect” occurred inbehavior diffusion. Then, we use both the artificially constructed social network graphs (with thefeatures of power-law degree distribution and small-world characteristics) and the real-data tracesof social networks to verify the performance of our proposal (in terms of the eventually influencedusers with the number of initial seeds). The simulations illustrate that our proposed Pagerank-likealgorithm can advantage over the existing degree-based discount algorithm (DegreeDiscount), andachieve the comparable performance as greedy algorithm.Considering the fact that, in our everyday social life, there exist many scenarios that users areaffected by each other, and decisions are made collaboratively, we deeply investigate the behaviordiffusion model and diffusion process in social network, based on mean-field theory. In detail, wefirstly design a specific diffusion model that can accommodate the impact of three factors on user’sprobability of adopting one specific behavior: the absolute number and the ratio of neighborsadopting the behavior and the number of her total neighbors. Then, based on mean-field theory, thediffusion process is analytically provided, and the effect of network structure on behavior diffusionis also given. Finally, the necessary condition (critical threshold) for widespread diffusion isanalyzed. The simulation results verify the theoretical analysis.