知识表示学习常用数据集

dataset	#relation	#entity	# triple(train/valild/test)
WN11	11	38696	112581     2609    10544
WN18	18	40943	141442     5000     5000
FB13	13	75043	316232     5908     23733
FB15K	1345	14951	483142     50000     59071
FB1M	23382	1*10^6	17.5*10^6    50000     177404
FB5M	1192	5385322	19193556     5000     59071

WN11

• 出处（SE）

Bordes A, Weston J, Collobert R, et al. Learning Structured Embeddings of Knowledge Bases[C]//AAAI. 2011, 6(1): 6. PDF

• 特点

• As WordNet is composed of words with different meanings, here we term its entities as the concatenation of the word and an number indicating which sense it refers to i.e. auto_1 is the entityencoding the first meaning of the word “auto”.
• 举例：(_auto_1, _has_instance, _s_u_v_1)

WN18

数据集

• 出处（SME）

Bordes A, Glorot X, Weston J, et al. A semantic matching energy function for learning with multi-relational data[J]. Machine Learning, 2014, 94(2): 233-259. PDF

• 特点

• entities (termed synsets) correspond to senses, and relation types define lexical relations between those senses.
• As WordNet is composed of words with different meanings, we describe its entities by the concatenation of the word, its part-of-speech tag (‘NN’ for noun, ‘VB’ for verb, ‘JJ’ for adjective and ‘RB’ for adverb) and a digit indicating which sense it refers to i.e. _score_NN_1 is the entity encoding the first meaning of the noun “score”. This version of WordNet is different from that used in Bordes et al. (2011) because the original data has been preprocessed differently: this version contains less entities but more relation types.
• 举例： (_score_NN_1, _hypernym, _evaluation_NN_1)

FB13

• 出处（SE）

Bordes A, Weston J, Collobert R, et al. Learning Structured Embeddings of Knowledge Bases[C]//AAAI. 2011, 6(1): 6. PDF

• 特点

举例：(_marylin_monroe, _profession, _actress)

FB15K

数据集

• 出处（TransE）

Bordes A, Usunier N, Garcia-Duran A, et al. Translating embeddings for modeling multi-relational data[C]//Advances in neural information processing systems. 2013: 2787-2795. PDF

• 特点

• To make a small data set to experiment on we selected the subset of entities that are also present in the Wikilinks database and that also have at least 100 mentions in Freebase (for both entities and relationships). We also removed relationships like ’!/people/person/nationality’ which just reverses the head and tail compared to the relationship ’/people/person/nationality’.

FB1M

• 出处（TransE）

Bordes A, Usunier N, Garcia-Duran A, et al. Translating embeddings for modeling multi-relational data[C]//Advances in neural information processing systems. 2013: 2787-2795. PDF

• 特点

We also wanted to have large-scale data in order to test TransE at scale. Hence, we created another data set from Freebase, by selecting the most frequently occurring 1 million entities.