Knowledge Graph Completion with Mixed Geometry Tensor Factorization

Viacheslav Yusupov, Maxim Rakhuba, Evgeny Frolov
Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, PMLR 258:4924-4932, 2025.

Abstract

In this paper, we propose a new geometric approach for knowledge graph completion via low rank tensor approximation. We augment a pretrained and well-established Euclidean model based on a Tucker tensor decomposition with a novel hyperbolic interaction term. This correction enables more nuanced capturing of distributional properties in data better aligned with real-world knowledge graphs. By combining two geometries together, our approach improves expressivity of the resulting model achieving new state-of-the-art link prediction accuracy with a significantly lower number of parameters compared to the previous Euclidean and hyperbolic models.

Cite this Paper

BibTeX
@InProceedings{pmlr-v258-yusupov25a, title = {Knowledge Graph Completion with Mixed Geometry Tensor Factorization}, author = {Yusupov, Viacheslav and Rakhuba, Maxim and Frolov, Evgeny}, booktitle = {Proceedings of The 28th International Conference on Artificial Intelligence and Statistics}, pages = {4924--4932}, year = {2025}, editor = {Li, Yingzhen and Mandt, Stephan and Agrawal, Shipra and Khan, Emtiyaz}, volume = {258}, series = {Proceedings of Machine Learning Research}, month = {03--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v258/main/assets/yusupov25a/yusupov25a.pdf}, url = {https://proceedings.mlr.press/v258/yusupov25a.html}, abstract = {In this paper, we propose a new geometric approach for knowledge graph completion via low rank tensor approximation. We augment a pretrained and well-established Euclidean model based on a Tucker tensor decomposition with a novel hyperbolic interaction term. This correction enables more nuanced capturing of distributional properties in data better aligned with real-world knowledge graphs. By combining two geometries together, our approach improves expressivity of the resulting model achieving new state-of-the-art link prediction accuracy with a significantly lower number of parameters compared to the previous Euclidean and hyperbolic models.} }
Endnote
%0 Conference Paper %T Knowledge Graph Completion with Mixed Geometry Tensor Factorization %A Viacheslav Yusupov %A Maxim Rakhuba %A Evgeny Frolov %B Proceedings of The 28th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2025 %E Yingzhen Li %E Stephan Mandt %E Shipra Agrawal %E Emtiyaz Khan %F pmlr-v258-yusupov25a %I PMLR %P 4924--4932 %U https://proceedings.mlr.press/v258/yusupov25a.html %V 258 %X In this paper, we propose a new geometric approach for knowledge graph completion via low rank tensor approximation. We augment a pretrained and well-established Euclidean model based on a Tucker tensor decomposition with a novel hyperbolic interaction term. This correction enables more nuanced capturing of distributional properties in data better aligned with real-world knowledge graphs. By combining two geometries together, our approach improves expressivity of the resulting model achieving new state-of-the-art link prediction accuracy with a significantly lower number of parameters compared to the previous Euclidean and hyperbolic models.
APA
Yusupov, V., Rakhuba, M. & Frolov, E.. (2025). Knowledge Graph Completion with Mixed Geometry Tensor Factorization. Proceedings of The 28th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 258:4924-4932 Available from https://proceedings.mlr.press/v258/yusupov25a.html.

Related Material