Abstract

People use online social networks to exchange information, spread ideas, learn about innovations, etc. Thus, it is important to know how information spreads through social networks. It is possible to spread information (e.g., product advertisement) to a larger number of individuals via a social network. Similarly, it is possible to minimize the spread of unwanted content (e.g., ‘false news’). The key point in both cases is to identify the most influential individuals on the social network. This problem is named as Influence Maximization (IM) problem. The IM problem focuses on finding the small subset of individuals in a social environment who influence a certain group of individuals, i.e., maximize/minimize the spreading of information. Some greedy algorithms, stochastic algorithms and evolutionary optimization algorithms have been developed to find a solution to this problem. However, these methods are not at the desired level in terms of speed or solution capability. On the other hand, although many swarm intelligence algorithms that produce fast and optimal solutions can be found in the literature, these algorithms cannot be directly applied to the IM problem because no general slope is produced on the state-space surface of the IM problem's objective function. Swarm intelligence algorithms follow the general slope over the surface to converge at the global optimum. Thus, they cannot converge to the global optimum in the IM problem.In this study, a change in the structure of the IM problem is suggested in order to tailor it to swarm intelligence algorithms and to achieve a general slope on the state-space surface of its objective function. We named this process as “reshaping”. More precisely, if a social network is envisioned as a graph and individuals as nodes, reshaping means sorting the nodes in descending order (from largest to smallest) according to the metrics under consideration (i.e., metrics that give an idea about the level of influence of an individual) and renumbering the nodes according to this order. Thus, the nodes those are close to each other in terms of level of influence become closer to each other in the state-space. This creates a general slope on the state-space surface of the objective function. This simple idea paves the way for applying all swarm intelligence algorithms to this kind of problem. The proposed approach was tested with real and synthetic graphs. The experiments employed the Grey Wolf Optimizer (GWO) and Whale Optimization Algorithm (WOA) as the swarm intelligence algorithms and PageRank and Kempe et al.’s Greedy Algorithm as benchmark methods. Experimental results showed that this approach worked well.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call