TAHB: New Benchmark for Text-Attributed Hypergraph Learning
A team of researchers has unveiled the TAHB, or Text-Attributed Hypergraph Benchmark, marking a groundbreaking integration of hypergraph structures with textual features. This innovation allows for the representation of complex relationships beyond binary ones. The TAHB fills a notable void in the benchmarking landscape by offering a comprehensive evaluation platform comprised of 10 datasets spanning four domains: e-commerce, academia, film, and political networks. Initial findings indicate that TAHB effectively maintains the vital characteristics of authentic hypergraphs and reflects the performance trends seen in existing benchmarks. The research paper detailing this development is available on arXiv under ID 2608.15055.
Key facts
- TAHB is the first public benchmark integrating hypergraph structures and raw textual attributes.
- It contains 10 real-world datasets from four domains: e-commerce, academia, movies, and politics networks.
- The benchmark enables systematic evaluation of text-aware hypergraph representation learning.
- Experimental results show TAHB preserves key structural properties of real-world hypergraphs.
- TAHB consistently reproduces performance tendencies observed in existing benchmarks.
- The research addresses the lack of public text-attributed hypergraph benchmarks.
- The paper is available on arXiv with identifier 2608.15055.
- Hypergraphs model higher-order groupwise relationships beyond pairwise interactions.
Entities
Institutions
- arXiv