Skip to content

Latest commit

 

History

History
62 lines (47 loc) · 3.76 KB

README.md

File metadata and controls

62 lines (47 loc) · 3.76 KB
Raw

WD50k dataset: An hyper-relational dataset derived from Wikidata statements

The dataset is constructed by the following procedure based on the Wikidata RDF dump of August 2019:

  • A set of seed nodes corresponding to entities from FB15K-237 having a direct mapping in Wikidata (P646 "Freebase ID") is extracted from the dump.
  • For each seed node, all statements whose main object and qualifier values corresponding to wikibase:Item are extracted from the dump.
  • All literals are filtered out from the qualifiers of the above obtained statements.
  • All the entities from the dataset which have less than two mentions are dropped. The statements corresponding to the dropped entities are also dropped.
  • The remaining statements are randomly split into the train, test, and validation sets.
  • All statements from train and validation sets are removed which share the same main triple (s,p,o) with test statements.
  • WD50k_33, WD50k_66, WD50k_100 are then sampled from the above statements. Here 33, 66, 100 represents the amount of hyper-relational facts (statements with qualifiers) in the dataset.

The table below provides some basic statistics of our dataset and its three further variations:

Dataset Statements w/Quals (%) Entities Relations E only in Quals R only in Quals Train Valid Test
WD50K 236,507 32,167 (13.6%) 47,156 532 5460 45 166,435 23,913 46,159
WD50K (33) 102,107 31,866 (31.2%) 38,124 475 6463 47 73,406 10,668 18,133
WD50K (66) 49,167 31,696 (64.5%) 27,347 494 7167 53 35,968 5,154 8,045
WD50K (100) 31,314 31,314 (100%) 18,792 279 7862 75 22,738 3,279 5,297

Each dataset i.e. wd50k and its derivatives (for example wd_50k) consists of two folders

  • statements which corresponds to dataset with qualifiers.
  • triples which corresponds to dataset where all qualifiers have been removed. For example a fact in statements (s, r, o, {(qr_1, qv_1), (qr_2, qv_2)}) is reduced to (s, r, o,) in triples.

These folders have train.txt, test.txt and valid.txt corresponding to train, test, and valid splits. Each line in the text file represents a fact/statement in the format s, r, o, qr_1, qv_1, qr_2, qv_2 .... The first three elements (s,r,o) represents the main triple, and the remaining part is the qualifier information. Note that the fact might not contain qualifier information (qr_1, qv_1, qr_2, qv_2 ...). Below are a few examples of the dataset:

Q515632,P1196,Q3739104
Q219546,P1411,Q103916,P805,Q369706,P1686,Q3241699
Q131074,P166,Q487136,P805,Q458646,P1346,Q630767,P1346,Q15840165
Q965,P530,Q117,P805,Q2564434
Q825807,P1889,Q502273

In the above snippet for line Q219546,P1411,Q103916,P805,Q369706,P1686,Q3241699

Item Description
Q219546 subject (s)
P1411 relation (r)
Q103916 object (o)
P805 qualifier relation 1 (qr_1)
Q369706 qualifier entity 1 (qe_1)
P1686 qualifier relation 2 (qr_2)
Q3241699 qualifier entity 2 (qe_2)

When using the dataset please cite:

@inproceedings{StarE,
  title={Message Passing for Hyper-Relational Knowledge Graphs},
  author={Galkin, Mikhail and Trivedi, Priyansh and Maheshwari, Gaurav and Usbeck, Ricardo and Lehmann, Jens},
  booktitle={EMNLP},
  year={2020}
}

For any further questions, please contact: mikhail.galkin@iais.fraunhofer.de