DMHANT: DropMessage Hypergraph Attention Network for Information Propagation Prediction.

Area of Science:

• Social Network Analysis
• Information Propagation Dynamics
• Machine Learning for Network Science

Background:

• Predicting information cascades is vital for understanding social network dynamics.
• Existing models often overlook global user dependencies and lack robustness.
• Characterizing dynamic user interaction preferences remains a challenge.

Purpose of the Study:

• To propose a robust and accurate model for predicting information propagation cascades.
• To address limitations in existing methods regarding global dependencies and model robustness.
• To enhance the characterization of dynamic user interaction preferences in social networks.

Main Methods:

• Constructing a hypergraph from cascade sequences to capture global dependencies.
• Developing hypergraph attention networks with time-stamped subgraphs to learn user interactions.
• Implementing a gated fusion strategy and a novel DropMessage method for robustness.

Main Results:

• The proposed DropMessage Hypergraph Attention Networks significantly outperform state-of-the-art models in MAP@k and Hits@K metrics.
• The model demonstrates superior prediction performance compared to existing methods under data perturbation.
• Experimental validation on three real-world datasets confirms the model's effectiveness.

Conclusions:

• The DropMessage Hypergraph Attention Networks offer a significant advancement in information cascade prediction.
• The model effectively captures global dependencies and enhances robustness through novel techniques.
• This approach provides a more accurate and reliable method for analyzing information propagation in social networks.