ENIMNR: Enhanced node influence maximization through node representation in social networks

Abstract

The influence maximization problem grapples with issues such as low infection rates and high time complexity. Many existing methods prove unsuitable for large-scale networks due to their time complexity or heavy reliance on free parameters. This paper introduces a solution to these challenges through a local heuristic that incorporates shell decomposition, node representation. This strategic approach selects candidate nodes based on their connections within network shells and topological features, effectively reducing the search space and computational overhead. The algorithm employs a deep learning-based node embedding technique to generate a low-dimensional vector for candidate nodes, calculating the dependency on spreading for each node based on local topological features. In the final phase, influential nodes are identified using results from previous phases and newly defined local features. Evaluation using the independent cascade model demonstrates the competitiveness of the proposed algorithm, highlighting its ability to deliver optimal performance in terms of solution quality. When compared to the Collective-Influence (CI) global algorithm, the presented method has a significant improvement in the differential infection rate due to its faster execution.