TensorSuite

Real Tensor

Higher-Arity Relational Knowledge Graphs

This TensorSuite collection includes element-wise tensors from higher-arity relational knowledge graph workloads, where relational facts involve more than two entities — also known as n-ary relational knowledge bases, multi-fold relational data, and hyper-relational knowledge graphs.

The tensors are sourced from two widely-used n-ary relational datasets: JF17K [1], derived from the popular knowledge graph Freebase, and WikiPeople [2]. Each dataset is represented at two arity levels, 3-ary and 4-ary, yielding four tensors in total.

Tensors

Dataset 3-ary 4-ary
JF17K JF17K-3 [Link]
order: 4, dim: 104×11541×11541×11541, nnz: 34,544		 JF17K-4 [Link]
order: 5, dim: 23×6536×6536×6536×6536, nnz: 9,509
WikiPeople WikiPeople-3 [Link]
order: 4, dim: 66×12270×12270×12270, nnz: 25,820		 WikiPeople-4 [Link]
order: 5, dim: 50×9528×9528×9528×9528, nnz: 15,188

References

[1] J. Wen, J. Li, Y. Mao, S. Chen, and R. Zhang, "On the Representation and Embedding of Knowledge Bases beyond Binary Relations," IJCAI, 2016. PDF

[2] S. Guan, X. Jin, Y. Wang, and X. Cheng, "Link Prediction on N-ary Relational Data," WWW, 2019. DOI