Novel graph neural networks using topological analysis improve performance in molecular prediction and social networks, indicating a new research direction.
Key Points
To introduce Topological Graph Neural Networks (TopGNNs) that enhance traditional graph neural networks using topological data analysis.
Proposed a novel framework incorporating topological data analysis into graph neural networks.
Utilized persistent homology and simplicial complexes to capture multi-scale structural information.
Conducted empirical analysis on benchmark datasets in various domains.
TopGNNs demonstrated competitive performance against state-of-the-art methods.
Achieved better sensitivity to global graph structure compared to conventional GNNs.