GNPA: a hybrid model for social influence maximization in dynamic networks

With the growing size of online social communities, influence propagation in social networks has become a hot topic of the research. Most of the studies in this area are based on the assumption that the structure of the social network is static and does not change during the information spread process. However, real-world social networks are dynamic. Modeling this continuous dynamic behavior of social networks is a challenge that must be tackled. This paper proposes a hybrid Genetic Network Programming with Apriori algorithm (GNPA) for Influence Maximization (IM) in social networks. Proposed GNPA is a meta-heuristic based optimization algorithm that handles the dynamicity, i.e. user attribute values or connection between users that changes with time. The working of GNPA is divided into 5 steps. It begins with the identification of the initial population of seeds using discounted degree method. Next, it predicts the future changes using the Apriori algorithm and updates the network accordingly. After updating the network dynamics, the influence score is estimated using the local consistent Factorization machines for each edge of the network. Finally, the diffusion score of each individual of the population is calculated using a linear cascade model. After the completion of all the above steps, the population is updated by replacing the weak individuals with new individuals using mutation and crossover, which is the last step of GNPA. The efficacy of GNPA is evaluated over two real and two synthetic datasets with low out-degree ratio and high out-degree ratio. Experimental results demonstrated that GNPA is able to predict the changing behavior of the users close to the actual-time network and improved the influence propagation 16% to 38% as compared to the contemporary counterparts.