Semi-HyperGraph Benchmark: Enhancing Flexibility of Hypergraph Learning with Datasets and Benchmarks

Zehui Li, Xiangyu Zhao, Mingzhu Shen, Guy-Bart Stan, Pietro Lio, Yiren Zhao

Keywords: Graph Neural Networks, Hypergraph Learning, Datasets and Benchmarks

TL;DR: We introduce a useful and flexible extension of hypergraphs by including simple edges, and present the Semi-HyperGraph Benchmark (SHGB), a collection of datasets combining hypergraphs and simple edges, with an extensible evaluation framework.

Abstract: Graphs are widely used to encapsulate a variety of data formats, but real-world networks often involve complex node relations beyond only being pairwise. While hypergraphs have been developed and employed to account for the complex node relations, they reduce the flexibility of machine learning systems by totally disregarding simple edges, which to some extent leads to a drop in performance. Additionally, Graph Neural Networks (GNNs) research are normally separated into simple graphs and hypergraphs, and these two classes of methods tend not to interchange. Therefore, there is a need for a more flexible benchmark that allows GNNs to employ both simple edge and hyperedge information. In this paper, we present the *Semi-HyperGraph Benchmark (SHGB)*, a collection of comprehensive datasets combining hypergraphs and simple edges, with an accessible evaluation framework to fully understand the performance of GNNs on complex graphs. SHGB contains 23 real-world hypergraph datasets with simple edges included, across various domains such as biology, social media, and e-commerce. Furthermore, we provide an extensible evaluation framework and a supporting codebase to facilitate the training and evaluation of GNNs on SHGB. Our empirical study of existing GNNs on SHGB reveals various research opportunities and gaps, including (1) evaluating the actual performance improvement of hypergraph GNNs over simple graph GNNs; (2) comparing the impact of different sampling strategies on hypergraph learning methods; and (3) exploring ways to integrate simple edge and hyperedge information. We make our source code and full datasets publicly available at https://anonymous-url/.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6950