Influence Maximization Model Based on Graph Comparison Learning

Influence maximization, as a key issue in social network analysis, aims to identify the most influential seed nodes to maximize the information propagation range. Traditional methods are mainly based on network topology, which suffer from the problems of low computational efficiency and underutilized node features. Therefore, this paper proposes a joint training framework GCL-IM that integrates graph contrast learning and multilayer perceptron to address the limitations of existing methods. First, the node representation vectors are constructed through a graph comparison learning framework to effectively capture the topological features and node attribute features of the network. Second, a regressor is designed to predict node importance to achieve collaborative modeling of topological features and attribute features. Finally, predicting network-wide node influence after training on a small-sized node set significantly reduces the algorithm training time while keeping the algorithm performance strongly competitive. The real network experiments show that GCL-IM has significant advantages over the benchmark algorithm in terms of propagation range and computational efficiency.