Modeling Controlled Information Diffusion In Online Social Networks

Social influence is constantly exploited by applications like online mar-keting to initiate chain-like cascades for disseminating messages. The diffu-sion is therefore regulated by the self-organized network structures and sponta-neous social interactions among users. However, recent practices gained more control of the diffusion process, which transforms the conventional diffusion paradigm into a new one, namely controlled diffusion. By creating supporting structures, they managed to incorporate massive collaborations among agents into the spreading process to inflate the pairwise activation probabilities and thus alter the spontaneous social dynamics and resulting influence.This paper proposed a framework for characterizing the underlying mech-anism of controlled diffusion and modeling its dynamics and influences. An-alytical and simulative models were derived to capture the overall and micro diffusion dynamics, respectively. Model analytics and simulations on a large Twitter network both confirmed the model conjecture on influence inflations. As to the maximization of the resulting influence, this work presented a two-step approach by introducing a greedy agents-selection procedure based on submod-ular property and optimizing their supporting structures under KKT conditions. Substantial simulations on five typical networks indicate that the performance of different optimization strategies depends on the very construction of network structures. As empirical study, the mining of a Weibo network revealed well-organized marketing groups which succeeded in prominently amplifying the influence of commercials via elaborate core-peripheral structures.