Royalty-20k and Royalty-30k datasets

Loading the data:

import utils
import numpy as np

DATASET = 'royalty_30k'
RULE = 'spouse'

data = np.load(DATASET+'.npz')

triples,traces,entities,relations = utils.get_data(data, RULE)

DATASET can be either 'royalty_30k' or 'royalty_20k'

For Royalty-30k, RULE can be 'spouse', 'successor', 'predecessor' or 'full_data'

For Royalty-20k, RULE can be 'spouse', 'grandparent' or 'full_data'

triples is a (N,3) numpy array containing the triples used for link prediction, with N being the total number of triples

traces is a (N,2,3) numpy array containing the explanations for each triple in triples

triples[0] is a triple we want an explanation for, traces[0] gives the only explanation for why triples[0] is a fact. There can be more than one triple in the explanation, for these datasets, the maximum is 2.

entities is a numpy array of all unique entities

relations is a numpy array of all unique relations

Linked Data Ground Truth for Quantitative and Qualitative Evaluation of Explanations for Relational Graph Convolutional Network Link Prediction on Knowledge Graphs

To reproduce the results from this paper, navigate to the /reproduce_results/ directory,

then build a conda environment which uses Python 3.7 and Tensorflow-GPU 2.3:

conda env create -f kg_env.yml --name kg_env

For the most recent benchmark results, navigate to the /latest/ directory

Please use the following citations:

@inproceedings{halliwell:hal-03430113,
  TITLE = {{Linked Data Ground Truth for Quantitative and Qualitative Evaluation of Explanations for Relational Graph Convolutional Network Link Prediction on Knowledge Graphs}},
  AUTHOR = {Halliwell, Nicholas and Gandon, Fabien and Lecue, Freddy},
  URL = {https://hal.archives-ouvertes.fr/hal-03430113},
  BOOKTITLE = {{International Conference on Web Intelligence and Intelligent Agent Technology}},
  ADDRESS = {Melbourne, Australia},
  YEAR = {2021},
  MONTH = Dec,
  DOI = {10.1145/3486622.3493921},
  KEYWORDS = {link prediction ; Explainable AI ; knowledge graphs ; graph neural networks},
  PDF = {https://hal.archives-ouvertes.fr/hal-03430113v2/file/WI_IAT.pdf},
  HAL_ID = {hal-03430113},
  HAL_VERSION = {v2},
}

@misc{halliwell:hal-03339562,
  TITLE = {{A Simplified Benchmark for Non-ambiguous Explanations of Knowledge Graph Link Prediction using Relational Graph Convolutional Networks}},
  AUTHOR = {Halliwell, Nicholas and Gandon, Fabien and Lecue, Freddy},
  URL = {https://hal.archives-ouvertes.fr/hal-03339562},
  NOTE = {Poster},
  HOWPUBLISHED = {{International Semantic Web Conference}},
  YEAR = {2021},
  MONTH = Oct,
  KEYWORDS = {knowledge graphs ; Explainable AI ; link prediction},
  PDF = {https://hal.archives-ouvertes.fr/hal-03339562v2/file/paper326.pdf},
  HAL_ID = {hal-03339562},
  HAL_VERSION = {v2},
}