freebase_fb15k — Freebase knowledge graph benchmarks (FB15k and FB15k-237, 2013-2015)
- Description
Two subsets of the Freebase knowledge base, widely used as benchmarks for knowledge graph completion (link prediction):
fb15k: the FB15k dataset of Bordes et al. (2013), with the 14,951 entities of Freebase that are also present in the Wikilinks database and have at least 100 mentions in Freebase, and the 1,345 relations between them that have at least 100 instances (592,213 triples); relations that merely reverse another one (e.g. the reverse of/people/person/nationality) were removed.fb15k_237: the FB15k-237 dataset of Toutanova and Chen (2015), a subset of FB15k from which near-duplicate and inverse relations were removed (as well as the triples of the validation and test sets that link entities directly linked in the training set), keeping 237 relations and 14,541 entities (310,116 triples). The textual mentions of pairs of entities in the ClueWeb12 corpus, distributed with this dataset, are left out.
Each node is an entity (e.g. a person, a place, a film or a profession), identified by its Freebase machine identifier (mid); the data do not include the names of the entities. A directed edge from entity A to entity B is a triple (A, relation, B), e.g. that a person has a given profession. Some relations of Freebase are two-hop relations, in which an intermediate entity (a "compound value type") was collapsed out (e.g.
/film/film/release_date_s./film/film_regional_release_date/film_release_region). Pairs of entities may be connected by several edges with different relations.The triples are divided into training, validation and test sets by the benchmarks; all of them are included, and the set is given as an edge property.
The networks have the following properties:
mid: the Freebase identifier of the entity, e.g./m/027rn.- Edge
relation: the Freebase relation, e.g./people/person/profession. - Edge
split: the set of the benchmark the triple belongs to:train,validortest.
- Tags
- Informational Relatedness Unweighted Multigraph Metadata
- Citation
-
- A. Bordes, N. Usunier, A. Garcia-Durán, J. Weston, O. Yakhnenko, "Translating Embeddings for Modeling Multi-relational Data." Advances in Neural Information Processing Systems 26, 2787-2795 (2013) papers.nips.cc
- K. Toutanova, D. Chen, "Observed versus latent features for knowledge base and text inference." Proceedings of the 3rd Workshop on Continuous Vector Space Models and their Compositionality, 57-66 (2015) 10.18653/v1/W15-4007 Sci-Hub
- K. Toutanova, D. Chen, P. Pantel, H. Poon, P. Choudhury, M. Gamon, "Representing Text for Joint Embedding of Text and Knowledge Bases." Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, 1499-1509 (2015) 10.18653/v1/D15-1174 Sci-Hub
- Upstream URL OK
-
https://web.archive.org/web/20221204060113/https://everest.hds.utc.fr/doku.php?id=en:transe
https://www.microsoft.com/en-us/download/details.aspx?id=52312 - Upstream license
- CC BY 2.5 (fb15k); Microsoft Research Data License Agreement, for non-commercial or research purposes only (fb15k_237)
- Networks
-
Tip: click on a column header to sort the list, or hover over it for a description.
Column legend
- $\left<k\right>$
- Mean degree
- $\sigma_k$
- Standard deviation of the degrees
- $\lambda_h$
- Largest eigenvalue of the Hashimoto (non-backtracking) matrix
- $\tau$
- Random walk mixing time
- $r$
- Degree assortativity
- $c$
- Global clustering coefficient
- $\oslash$
- (Pseudo-) diameter
- $S$
- Relative size of the largest component
- NPs
- Node properties
- EPs
- Edge properties
Name Nodes Edges $\left<k\right>$ $\sigma_k$ $\lambda_h$ $\tau$ $r$ $c$ $\oslash$ $S$ Kind Mode NPs EPs gt GraphML GML csv fb15k 14,951 592,213 39.61 220.72 152.99 32.04 -0.11 0.03 8 1.00 Directed Unipartite mid relation split 2.0 MiB 4.0 MiB 3.9 MiB 4.0 MiB fb15k_237 14,541 310,116 21.33 127.70 149.27 447.20 -0.09 0.03 8 1.00 Directed Unipartite mid relation split 1.1 MiB 2.2 MiB 2.2 MiB 2.2 MiB - Ridiculograms*