همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Community-Aware Influence Maximization on Heterogeneous Information Networks using Graph Neural Networks

نویسندگان :

Amirreza Vafaei¹ Sadegh Aliakbary²

1- دانشگاه شهید بهشتی 2- دانشگاه شهید بهشتی

کلمات کلیدی :

Influence Maximization،Heterogeneous Networks،Community Detection،Graph Neural Networks

چکیده :

Abstract—Influence Maximization (IM) is a critical problem in network analysis, aiming to find a minimum set of seed nodes that maximizes information spread. While recent deep learning-based approaches, such as MAHE-IM, have shown success in Heterogeneous Information Networks (HINs), they often rely on less powerful embedding techniques like Skip-gram (e.g., metapath2vec) and fail to explicitly utilize the network's inherent community structure. To address these limitations, we propose GNN-IM-CD (Graph Neural Network for Influence Maximization based on Community Detection). This Algorithm introduces two main enhancements: 1) replacing the Skip-gram component with a powerful Heterogeneous Graph Neural Network (H-GNN) to learn richer node representations; and 2) integrating an Enhanced Community Detection (CD) layer using an aggressive Louvain method to enforce seed diversity and leverage sub-structure knowledge. Experimental results on real-world HIN datasets demonstrate that GNN-IM-CD significantly outperforms the state-of-the-art MAHE-IM framework under both the Independent Cascade (IC) and Linear Threshold (LT) models, validating the power of combining GNNs with community structure awareness for effective influence spread.