Scholar Influence Maximization via Opinion Leader and Graph Embedding Regression in Social Networks

The significance of influential nodes in information diffusion within scholar social networks has attracted increasing attention in the field of influence maximization problem. Most Existing influence maximization algorithms relying on scenario-specific centrality measures often underperform when applied to diverse social networks. Many influence maximization deep learning based algorithms use topological embedding to learn the global influence information. However, since most of these methods primarily focus on identifying high influence nodes to solve the influence maximization problem, they often suffer from poor performance due to the overlapping influence ranges among highly influential nodes within the seed set. In this paper, we propose a regression framework via opinion leaders and graph embedding, named Inf-LGR, to optimize the performance degradation problem caused by the highly overlapping influence ranges between the highly influential nodes in the seed set. Specifically, our framework first adopts an information diffusion model-based approach to obtain opinion leaders’ tendency evaluation and meanwhile uses variational graph auto-encoders (VGAE) to encode it into low-dimensional vectors. Then we utilize an improved graph sample and aggregate (GraphSAGE) algorithm, whose aggregation based on the number of shortest paths, to generate the influence embedding of nodes. Finally, embedding results are fed into a regressor for predicting the influence score of each node. The top-K nodes with the highest scores are selected as the seed set of the influence maximization problem. Experimental results on four social networks demonstrate the proposed framework outperforms some of the recently proposed and classical influence maximization methods.