Learning Node Embeddings for Influence Set Completion

Sergey Ivanov, Nikita Durasov, Evgeny Burnaev

Abstract: Influence Maximization problem has found numerous applications in the real world and attracted a lot of research in the recent years. However, all previous attempts to provide a solution were based solely on the graph topology. Instead, we show how to employ recent advancement in representation learning and use node embeddings for finding solution as a supervised task. In our experiments, we show that the ranked list of nodes obtained by classifier yields better influence completion set than other baselines.