metadata
language:
- en
license:
- other
multilinguality:
- monolingual
size_categories:
- n<1K
pretty_name: link_prediction_nell_one
Dataset Card for "relbert/link_prediction_nell_one"
Dataset Description
- Repository: https://github.com/xwhan/One-shot-Relational-Learning
- Paper: https://aclanthology.org/D18-1223/
- Dataset: Few-shots link prediction
Dataset Summary
This is NELL-ONE/Wiki-One dataset for the few-shots link prediction proposed in https://aclanthology.org/D18-1223/.
| train | validation | test | |
|---|---|---|---|
| nell | 8,526 | 1,004 | 2,158 |
| nell_filter | 5,498 | 878 | 1,352 |
TODO
- map the wikidata id to the name of the enity for the wiki-one dataset
Dataset Structure
Data Instances
An example of test of nell looks as follows.
{
"relation": "concept:sportsgamesport",
"head": "concept:sportsgame:n1937_world_series",
"tail": "concept:sport:baseball"
}
Statistics on the NELL test split
- Entity Types (nell)
| entity_type | tail | head |
|---|---|---|
| videogame | 0 | 4 |
| crustacean | 25 | 11 |
| organization | 2 | 32 |
| invertebrate | 43 | 14 |
| agriculturalproduct | 0 | 87 |
| astronaut | 1 | 0 |
| vegetable | 0 | 8 |
| geopoliticallocation | 14 | 24 |
| person | 96 | 14 |
| drug | 0 | 1 |
| arthropod | 41 | 32 |
| location | 2 | 0 |
| company | 147 | 1 |
| female | 3 | 3 |
| product | 0 | 62 |
| chemical | 0 | 1 |
| legume | 0 | 1 |
| county | 4 | 10 |
| mlsoftware | 0 | 1 |
| sport | 74 | 93 |
| fruit | 0 | 1 |
| personeurope | 1 | 0 |
| planet | 1 | 0 |
| mammal | 0 | 23 |
| professor | 0 | 1 |
| criminal | 1 | 0 |
| athlete | 59 | 34 |
| insect | 270 | 230 |
| school | 0 | 11 |
| automobilemodel | 0 | 100 |
| coffeedrink | 0 | 11 |
| food | 0 | 4 |
| celebrity | 2 | 4 |
| biotechcompany | 11 | 0 |
| animal | 30 | 97 |
| visualizablescene | 3 | 3 |
| politician | 58 | 23 |
| software | 0 | 42 |
| sportsgame | 0 | 74 |
| grain | 0 | 2 |
| city | 161 | 42 |
| personaustralia | 5 | 4 |
| politicianus | 360 | 352 |
| hobby | 0 | 10 |
| arachnid | 6 | 1 |
| country | 317 | 27 |
| male | 5 | 3 |
| personnorthamerica | 3 | 1 |
| sportsteam | 0 | 295 |
| stateorprovince | 0 | 38 |
| personus | 6 | 1 |
| coach | 245 | 1 |
| director | 1 | 0 |
| automobilemaker | 54 | 274 |
| reptile | 0 | 4 |
| journalist | 1 | 0 |
| vehicle | 0 | 2 |
| bodypart | 69 | 0 |
| amphibian | 0 | 2 |
| beverage | 0 | 27 |
| island | 0 | 1 |
| geopoliticalorganization | 17 | 1 |
| personmexico | 20 | 13 |
- Relation Types (nell)
| relation | relation |
|---|---|
| concept:agriculturalproductcamefromcountry | 140 |
| concept:animalsuchasinvertebrate | 415 |
| concept:athleteinjuredhisbodypart | 69 |
| concept:automobilemakerdealersincity | 178 |
| concept:automobilemakerdealersincountry | 96 |
| concept:geopoliticallocationresidenceofpersion | 143 |
| concept:politicianusendorsespoliticianus | 386 |
| concept:producedby | 213 |
| concept:sportschoolincountry | 103 |
| concept:sportsgamesport | 74 |
| concept:teamcoach | 341 |
Vocab Size (nell): 68544
Entity Types (nell_filter)
| entity_type | tail | head |
|---|---|---|
| videogame | 0 | 4 |
| organization | 2 | 32 |
| astronaut | 1 | 0 |
| geopoliticallocation | 8 | 24 |
| person | 96 | 0 |
| drug | 0 | 1 |
| company | 144 | 1 |
| female | 3 | 3 |
| product | 0 | 62 |
| county | 4 | 10 |
| personeurope | 1 | 0 |
| planet | 1 | 0 |
| professor | 0 | 1 |
| criminal | 1 | 0 |
| athlete | 59 | 0 |
| school | 0 | 11 |
| automobilemodel | 0 | 100 |
| celebrity | 2 | 4 |
| biotechcompany | 10 | 0 |
| visualizablescene | 3 | 3 |
| politician | 58 | 23 |
| software | 0 | 42 |
| city | 161 | 42 |
| personaustralia | 5 | 0 |
| politicianus | 360 | 352 |
| country | 91 | 27 |
| male | 5 | 1 |
| personnorthamerica | 3 | 0 |
| sportsteam | 0 | 295 |
| stateorprovince | 0 | 38 |
| personus | 6 | 1 |
| coach | 245 | 0 |
| director | 1 | 0 |
| automobilemaker | 54 | 273 |
| journalist | 1 | 0 |
| island | 0 | 1 |
| geopoliticalorganization | 7 | 1 |
| personmexico | 20 | 0 |
- Relation Types (nell_filter)
| relation | relation |
|---|---|
| concept:automobilemakerdealersincity | 177 |
| concept:automobilemakerdealersincountry | 96 |
| concept:geopoliticallocationresidenceofpersion | 143 |
| concept:politicianusendorsespoliticianus | 386 |
| concept:producedby | 209 |
| concept:teamcoach | 341 |
Vocab Size (nell_filter) 53887
Citation Information
@inproceedings{xiong-etal-2018-one,
title = "One-Shot Relational Learning for Knowledge Graphs",
author = "Xiong, Wenhan and
Yu, Mo and
Chang, Shiyu and
Guo, Xiaoxiao and
Wang, William Yang",
booktitle = "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing",
month = oct # "-" # nov,
year = "2018",
address = "Brussels, Belgium",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/D18-1223",
doi = "10.18653/v1/D18-1223",
pages = "1980--1990",
abstract = "Knowledge graphs (KG) are the key components of various natural language processing applications. To further expand KGs{'} coverage, previous studies on knowledge graph completion usually require a large number of positive examples for each relation. However, we observe long-tail relations are actually more common in KGs and those newly added relations often do not have many known triples for training. In this work, we aim at predicting new facts under a challenging setting where only one training instance is available. We propose a one-shot relational learning framework, which utilizes the knowledge distilled by embedding models and learns a matching metric by considering both the learned embeddings and one-hop graph structures. Empirically, our model yields considerable performance improvements over existing embedding models, and also eliminates the need of re-training the embedding models when dealing with newly added relations.",
}