Convolutional 2D Knowledge Graph Embeddings
Authors: Tim Dettmers, Pasquale Minervini, Pontus Stenetorp, Sebastian Riedel
AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this work we introduce Conv E, a multi-layer convolutional network model for link prediction, and report state-of-the-art results for several established datasets. We also show that the model is highly parameter efficient, yielding the same performance as Dist Mult and R-GCN with 8x and 17x fewer parameters. Analysis of our model suggests that it is particularly effective at modelling nodes with high indegree which are common in highlyconnected, complex knowledge graphs such as Freebase and YAGO3. [...] We then perform experiments on these robust datasets for our own and several previously proposed models, and find that Conv E achieves state-of-the-art Mean Reciprocal Rank across all datasets. |
| Researcher Affiliation | Academia | Tim Dettmers Università della Svizzera italiana tim.dettmers@gmail.com Pasquale Minervini, Pontus Stenetorp, Sebastian Riedel University College London {p.minervini,p.stenetorp,s.riedel}@cs.ucl.ac.uk |
| Pseudocode | No | The paper describes the model architecture but does not include any explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code for our model and experiments is made publicly available,2 as well as the code for replicating the Dist Mult results.3 2https://github.com/Tim Dettmers/Conv E 3https://github.com/uclmr/inferbeddings |
| Open Datasets | Yes | For evaluating our proposed model, we use a selection of link prediction datasets from the literature. WN18 (Bordes et al. 2013a) is a subset of Word Net [...] FB15k (Bordes et al. 2013a) is a subset of Freebase [...] YAGO3-10 (Mahdisoltani, Biega, and Suchanek 2015) is a subset of YAGO3 [...] Countries (Bouchard, Singh, and Trouillon 2015) is a benchmark dataset [...] To create a dataset without this property, Toutanova and Chen (2015) introduced FB15k-237 a subset of FB15k where inverse relations are removed. [...] We thus create WN18RR to reclaim WN18 as a dataset, which cannot easily be completed using a single rule but requires modelling of the complete knowledge graph. WN18RR1 contains 93,003 triples with 40,943 entities and 11 relations. 1https://github.com/Tim Dettmers/Conv E |
| Dataset Splits | Yes | We selected the hyperparameters of our Conv E model via grid search according to the mean reciprocal rank (MRR) on the validation set. Hyperparameter ranges for the grid search were as follows embedding dropout {0.0, 0.1, 0.2}, feature map dropout {0.0, 0.1, 0.2, 0.3}, projection layer dropout {0.0, 0.1, 0.3, 0.5}, embedding size {100, 200}, batch size {64, 128, 256}, learning rate {0.001, 0.003}, and label smoothing {0.0, 0.1, 0.2, 0.3}. [...] We use early stopping according to the mean reciprocal rank (WN18, FB15k, YAGO3-10) and AUC-PR (Countries) statistics on the validation set, which we evaluate every three epochs. |
| Hardware Specification | No | The paper mentions running experiments on a 'high-end GPU' but does not specify any particular hardware model (e.g., specific GPU or CPU). |
| Software Dependencies | No | The paper mentions using 'rectified linear units', 'batch normalisation', 'dropout', and 'Adam as optimiser', but does not provide specific version numbers for any software libraries or frameworks used (e.g., PyTorch, TensorFlow, Python version). |
| Experiment Setup | Yes | We selected the hyperparameters of our Conv E model via grid search according to the mean reciprocal rank (MRR) on the validation set. Hyperparameter ranges for the grid search were as follows embedding dropout {0.0, 0.1, 0.2}, feature map dropout {0.0, 0.1, 0.2, 0.3}, projection layer dropout {0.0, 0.1, 0.3, 0.5}, embedding size {100, 200}, batch size {64, 128, 256}, learning rate {0.001, 0.003}, and label smoothing {0.0, 0.1, 0.2, 0.3}. [...] We found that the following combination of parameters works well on WN18, YAGO3-10 and FB15k: embedding dropout 0.2, feature map dropout 0.2, projection layer dropout 0.3, embedding size 200, batch size 128, learning rate 0.001, and label smoothing 0.1. For the Countries dataset, we increase embedding dropout to 0.3, hidden dropout to 0.5, and set label smoothing to 0. |