Finding influentials in social networks using evolutionary algorithm

The social influence maximization problem is an important research topic for many years since it has a tremendous impact on society. As social influence can be maximized for many purposes, such as marketing, politics, spreading innovations, there are many stakeholders interested in progress in this area. As it has been shown, for most settings finding an optimal seed set is an NP-hard problem, this is why various heuristics started to emerge since the statement of the problem. In this work, we explore the applicability of evolutionary algorithm for influence maximization. The experiments conducted using real and artificial social networks and the linear threshold influence model show that this approach offers not only speed and accuracy. Also, some other interesting features have been found, such as the transferability of its parameters to other datasets. Summarizing the results, it was observed that the evolutionary algorithm does not lose its performance when limiting its time by the factor of two and for most datasets, we obtained a high correlation of ranked parameters’ sets used for the evolutionary algorithm, typically around 0.9. Overall, these features combined make this approach an interesting research direction for influence maximization.