ContentSide / French_Knowledge_Graph

This repository introduces RezoJDM16K a French Knowledge Graph Dataset with 53 semantic relations created from RezoJDM. Different graph embeddings have gained from this dataset which are available for semantic link prediction task.

RezoJDM16k (A French Knowledge Graph DataSet)

This repository introduces RezoJDM16K a French Knowledge Graph Dataset with 53 semantic relations created from RezoJDM. Different graph embeddings have been gained from this dataset which are available for semantic link prediction tasks.

Instruction:

Simply, download the ./benchmarks/RezoJDM16K folder and use it for Knowledge Graph Embedding Models available in OpenKE. Run Jupiter notebook ./src/4-OpenKE_2_KGE_Models.ipynb which is created for this purpose.

Description of Source Notebooks

The Jupiter notebooks in ./src are for recreation the dataset with RezoJDM dump:

• 1-Dump2CSVs.ipynb : converts RezoJDM dumps to CSV format files with simple preliminary filters.
• 2-CSV2Triplets.ipynb : converts CSV format files into triplets.
• 3-Triplets2OpenKE.ipynb : convert triplets to OpenKE graph format.
• 4-OpenKE_2_KGE_Models.ipynb : use OpenKE graphs for trainnig with different knowledge graph embedding models.

Performance:

Here is the performance of knowledge graph embedding models for RezoJDM16k considering the different evaluation metrics [1]

Model	MRR	MR	Hits@10	Hits@3	Hits@1
TransE	0.179	203.31	0.432	0.242	0.041
TransH	0.218	177.12	0.498	0.291	0.069
TransD	0.216	170.68	0.500	0.287	0.066
DistMult	0.220	194.47	0.445	0.252	0.109
ComplEx	0.253	201.58	0.533	0.304	0.117

Here is the performance (Hits@10) of the different Knowledge Graph Embeddings Models run on RezoJDM16K compared to other English datasets.

Model	RezoJDM16k	WN18RR	FB15k237	WN18RR [2]	FB15k237 [2]
TransE	0.422	0.512	0.476	0.501	0.486
TransH	0.474	0.507	0.490	-	-
TransD	0.470	0.508	0.487	-	-
DistMult	0.424	0.479	0.419	0.49	0.419
ComplEx	0.528	0.485	0.426	0.510	0.428
RotatE *	0.583	0.565	0.522	0.571	0.533

RezoJDM16K Dataset Citation

@inproceedings{mirzapour2022,
  title={Introducing RezoJDM Knowledge Graph DataSet for Link Prediction},
  author={Mehdi Mirzapour, Waleed Ragheb, Mohammad Javad Saeedizade,Kevin Cousot, Helene Jacquenet, Lawrence Carbon, Mathieu Lafourcade},
  booktitle={({{To}} Appear) Proceedings of the 13th Language Resources and Evaluation Conference (LREC)},
  year={2022}
}

RezoJDM Citations

@inproceedings{lafourcade2007making,
  title={Making people play for Lexical Acquisition with the JeuxDeMots prototype},
  author={Lafourcade, Mathieu},
  booktitle={SNLP'07: 7th international symposium on natural language processing},
  pages={7},
  year={2007}
}

License

RezoJDM16K is released under the MIT license.

---

* These results were not reported in the original paper