Influence Maximization of Social Networks Using DRL

Influence Maximization can be defined as identifying highly influential individuals in a social network in such a strategic way that it achieves the maximum coverage of information spreading. The problem with existing methods is that they consist of computationally intensive strategies involving greedy algorithms or are simplistic representations of reality. Recently, advancements in graph representation learning and decision-making have emerged as effective alternatives for existing methods. The current study proposes utilizing GNNs coupled with DRL for influence maximization. Rather it should learn from the structure of the network itself how it can select the seed set of users who have the largest effect. The authors applied their framework on real-world networks such as Facebook and Epinions, sourced from the SNAP dataset, employing the Independent Cascade information spread process on a diffusion model. The results are encouraging. GNN-DRL provides better performance compared to other simplistic approaches, including Degree Centrality. It does so in a more computationally efficient manner and also shows the ability to be scalable and flexible. The most noticeable aspect of this particular method would be its ability to learn the actual spread of influence in the network as it gains experience from the network itself. This work will not only make influence maximization more effective but also open up broad future directions of work on adaptive graphbased optimization techniques.