KGATE : a Knowledge Graph Embedding Training Environment

Benjamin Loire; Galadriel Brière; Célia Brahimi; Antoine Toffano; Anaïs Baudot

cs.LGOct 7, 2026

KGATE : a Knowledge Graph Embedding Training Environment

Authors: Benjamin Loire, Galadriel Brière, Célia Brahimi, Antoine Toffano, Anaïs Baudot

Organizations: Aix Marseille Univ., INSERM, Marseille Medical Genetics, Systems Biomedicine Team, Marseille, France · Neurology Therapeutic Area, R&D Servier Paris-Saclay Institut, Gif-sur-Yvette, France · LIRMM, Univ. Montpellier, CNRS, Montpellier, France · CNRS, Marseille, France

Abstract

Knowledge graph embedding (KGE) models encode the entities and relations of a knowledge graph into a low-dimensional latent space, enabling tasks such as classification or link prediction. Most KGE models follow an autoencoder architecture, in which an encoder projects the knowledge graph into the latent space and a decoder reconstruct it. Combining both encoder and decoder components is increasingly needed, yet existing libraries rarely support complete autoencoders, are often unmaintained, rely on undocumented default hyperparameters, and produce results that cannot be compared across libraries. Here we present KGATE (Knowledge Graph Autoencoder Training Environment), a modular Python library built on PyTorch Geometric and TorchKGE. KGATE lets users assemble initializers, encoders, decoders, losses, negative samplers, and evaluation metrics as building blocks, or plug in their own block. KGATE includes a preprocessing procedure that controls data leakage, a builtin training pipeline, and reproducibility by design. Benchmarks against six existing KGE libraries show that KGATE training time is comparable with the fastest libraries while offering a broader set of features.

Figures & tables

Figure 1: KGATE building blocks. The numbers given for each building block are subject to change as the library evolves. The modular architecture allows for additional blocks to be included by users or future updates.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Supplementary Figure S1: Evolution of the embeddings of one gene node across 100 epochs using the training loops of five different libraries on the PrimeKG dataset. The embeddings were initialized once before being fed to each library as initial features to the nodes. Each run was performed on a single-threaded CPU to ensure determinism. A) Evolution of the two first components of a PCA performed on the embeddings across each epoch; B) Evolution of the L2 normalization of the embedding; C) Cosine similarity of the embedding at each epoch compared to the initial embedding.

Library	KGATE	TorchKGE	PyTorch Geometric	DGL-KE	Ampligraph	PyKEEN
KGATE	/	0.290	0.169	0.070	0.107	0.284
TorchKGE	0.290	/	0.260	0.058	0.077	0.197
PyTorch Geometric	0.0169	0.260	/	0.053	0.058	0.114
DGL-KE	0.070	0.058	0.053	/	0.053	0.064
Ampligraph	0.107	0.077	0.058	0.053	/	0.070
PyKEEN	0.284	0.197	0.114	0.064	0.070	/

Appendix

Supplementary Table S1: Mean Jaccard index of the similarity of the top 10 predictions for 10 different gene nodes sampled at random using the link prediction query “(gene, protein-protein, ?)”. Predictions are made once 100 epochs have elapsed, using the same model and the same hyperparameters when it is possible to tune them.

Library	Encoders	Decoders	KGE Autoencoder	Negative samplers	Data Leakage Control	Actively Maintained
KGATE (ours)	3	10	Yes	4	Advanced	Yes
PyTorch Geometric	21	4	No	1	Basic	Yes
TorchKGE	0	11	No	3	No	No
PyKEEN	3	34	One	3	Basic	Yes
LibKGE	0	11	No	2	No	No
Ampligraph	0	4	No	1	No	No

Appendix

Supplementary Table S2: Comparisons of KGATE and other KGE libraries utilities. The advanced data leakage of KGATE corresponds to the full procedure described in Brière et al, while Basic data leakage is a simple control of full duplicate edges. LibKGE appears to provide the necessary structure for an autoencoder; however no encoder is implemented. PyKEEN does implement a kind of autoencoder, with the implementation of specific message passing encoders like RGCN that take into account the different edge types. However, in PyKEEN, the generic message passing intended for user-driven customization treats the dataset as a homogeneous graph instead of a KG, failing to represent the complexity of the data.

Explore similar work

Pattern Over-Generalization of Knowledge Graph Embedding
Sep 3, 2026Junsik Kim, Kangil KimKnowledge Graph Embeddings Zero-Shot Link Prediction
Gwangju Institute of Science and Technology
GeoGAE: Scalable Graph-Level Autoencoding via Hyperball Cloud Representations
Sep 28, 2026Radosław Nowak, Anna Bielawska, Bogusz Stefańczyk +2Graph Representation Learning Graph Representations
Institute of Theoretical and Applied Informatics Polish Academy of Sciences · IDEAS Research Institute · Faculty of Mathematics, Informatics and Mechanics Warsaw University of Technology
Revisiting Catastrophic Forgetting in Continual Knowledge Graph Embedding
Apr 21, 2026Gerard Pons, Carlos Escolano, Besim Bilalli +1Knowledge Graph Embeddings Catastrophic Forgetting
Data-intensive Technologies and Knowledge Systems (DTAK), Department of Service and Information System Engineering, Universitat Politècnica de Catalunya, Barcelona, 08034, Catalonia, Spain · Intelligent Data Science and Artificial Intelligence Research Center (IDEAI), Department of Computer Science, Universitat Politècnica de Catalunya, Barcelona, 08034, Catalonia, Spain

Sep 3, 2026, cs.CLJ/K move · Enter open · S save

Pattern Over-Generalization of Knowledge Graph Embedding

Junsik Kim, Kangil Kim

Gwangju Institute of Science and Technology

Knowledge Graph Embeddings Zero-Shot Link Prediction Knowledge Graphs

Open paper AskPDF arXiv

Knowledge graph embedding (KGE) demonstrates its effectiveness for predicting missing links in knowledge graphs (KGs) by projecting entities and relations into a low-dimensional vector space. It is crucial for KGE models to effectively capture inference patterns (patterns) inherent in KGs, such as symmetry/antisymmetry, inversion and composition. Although recent KGE models exhibit strong capabilities in modeling such diverse patterns, they suffer from inherent limitations stemming from pattern over-generalization, where embeddings learned from only a single pattern instance inevitably generalize that pattern to all related instances, i.e., generalize the pattern universally. To address this issue, we propose PogRE (Pattern Over-Generalization Robust Embedding), a simple but effective method that utilizes dense linear transformations and compound operations for relation representation. Our theoretical analysis demonstrates that a dense linear transformation allows a pattern to become progressively universal as more triples are observed in the pattern. Furthermore, after observing d+1 linearly independent entities (d+1 denotes the dimension of entity), the linear transformation guarantees universal generalization of the pattern across all related instances. Experimental results on three standard benchmark datasets show that PogRE outperforms existing state-of-the-art KGE models in link prediction. Moreover, our empirical results indicate that PogRE effectively addresses the negative impact of over-generalization.

Abstract

Explore similar work

Pattern Over-Generalization of Knowledge Graph Embedding

GeoGAE: Scalable Graph-Level Autoencoding via Hyperball Cloud Representations

Revisiting Catastrophic Forgetting in Continual Knowledge Graph Embedding