Hypergraph Classification via Persistent Homology

Mehmet Emin Aktas,Thu Nguyen,Rakin Riza,Muhammad Ifte Islam,Esra Akbas

2023-06-20

Abstract:Persistent homology is a mathematical tool used for studying the shape of data by extracting its topological features. It has gained popularity in network science due to its applicability in various network mining problems, including clustering, graph classification, and graph neural networks. The definition of persistent homology for graphs is relatively straightforward, as graphs possess distinct intrinsic distances and a simplicial complex structure. However, hypergraphs present a challenge in preserving topological information since they may not have a simplicial complex structure. In this paper, we define several topological characterizations of hypergraphs in defining hypergraph persistent homology to prioritize different higher-order structures within hypergraphs. We further use these persistent homology filtrations in classifying four different real-world hypergraphs and compare their performance to the state-of-the-art graph neural network models. Experimental results demonstrate that persistent homology filtrations are effective in classifying hypergraphs and outperform the baseline models. To the best of our knowledge, this study represents the first systematic attempt to tackle the hypergraph classification problem using persistent homology.

Algebraic Topology,Social and Information Networks

What problem does this paper attempt to address?

The paper attempts to address the problem of applying Persistent Homology in hypergraph classification to extract topological features and evaluate its performance on different real-world datasets. Specifically:

1. **Defining Persistent Homology for Hypergraphs**: Traditional graphs have an obvious simplicial complex structure, making it intuitive to define persistent homology on them. However, hypergraphs lack this structure, making it challenging to preserve their topological information. The paper proposes several methods for defining topological features of hypergraphs to define persistent homology for hypergraphs, thereby prioritizing different higher-order structures within the hypergraph.

2. **Hypergraph Classification Performance**: By using these persistent homology filters to classify four different real-world hypergraphs and comparing their performance with current state-of-the-art graph neural network models. Experimental results show that persistent homology filters perform excellently in hypergraph classification tasks, outperforming baseline models.

3. **Comparison of Different Filters**: The study investigated three filtering methods—Simplicial Complex Closure (SCC), Restricted Barycentric Subdivision (ResBS), and Relative Barycentric Subdivision (RelBS), and conducted experimental validation on multiple datasets. The results show that the RelBS filtering method outperforms the other two methods in all dimensions.

Through this work, the study aims to fill the research gap in using persistent homology to address hypergraph classification problems and demonstrate the potential of this approach in practical applications.