This repository contains PyTorch implementation of RTuckER model for knowledge graph link prediction task.
The proposed method is a modification of the approach described in paper TuckER[1]. It represents the knowledge graph as a tensor with a fixed multilinear rank and uses the Riemannian optimization approach for training. Unlike TuckER model, this approach doesn't employ such common DL techniques as Dropout or BatchNormalization.
There are two types of models:
symmetric-- model uses equal subjects and objects embeddings in Tucker decomposition.asymmetric-- otherwise.
| Dataset | rank | lr | reg | batch size | decay rate | symmetric |
|---|---|---|---|---|---|---|
| WN18RR | (200, 20, 200) | 0.1 | 1e-10 | 2048 | 0.999 | True |
| FB15k-237 | (200, 200, 200) | 0.1 | 1e-10 | 512 | 0.999 | True |
| Dataset | MRR | Hits@10 | Hits@3 | Hits@1 |
|---|---|---|---|---|
| WN18RR | 0.449 | 0.508 | 0.468 | 0.414 |
| FB15k-237 | 0.313 | 0.481 | 0.342 | 0.231 |
TuckER: Tensor Factorization for Knowledge Graph Completion
MIT License