Hugging Face's logo

Datasets:
relbert
/
scientific_and_creative_analogy

Modalities:
Text
Languages:
English
Size:
< 1K
ArXiv:
Libraries:
Datasets
License:
Dataset card Viewer Files Community
1
scientific_and_creative_analogy / README.md
asahi417's picture
asahi417
init
6c1379c
preview code
|
raw
history blame contribute delete
No virus
2.7 kB
metadata
language:
  - en
license:
  - other
multilinguality:
  - monolingual
size_categories:
  - 1<n<1K
pretty_name: Relation Mapping

Dataset Card for "relbert/scientific_and_creative_analogy"

Dataset Description

Dataset Summary

A dataset for relation mapping task, which is a task to choose optimal combination of word pairs (see more detail in the paper).

Relation mapping M is the set of bijective map in between two sets of terms (A and B):

[set `A`]: ("solar system", "sun", "planet", "mass", "attracts", "revolves", "gravity")
[set `B`]: ("atom", "nucleus", "electron", "charge", "attracts", "revolves", "electromagnetism")

[Relation Mapping `M`]
* "solar system"   -> "atom"
* "sun"            -> "nucleus"
* "planet"         -> "electron"
* "mass"           -> "charge"
* "attracts"       -> "attracts"
* "revolves"       -> "revolves"
* "gravity"        -> "electromagnetism"

Relation Mapping Problem is the task to identify the mapping M given the sets of terms A and B.

Dataset Structure

Data Instances

An example looks as follows.

{
 "id": "0",
 "reference": ["buying an item", "accepting a belief"],
 "source": ["buying an item", "buyer", "merchandise", "buying", "selling", "returning", "valuable", "worthless"],
 "target": ["accepting a belief", "believer", "belief", "accepting", "advocating", "rejecting", "true", "false"],
 "target_random": ["rejecting", "true", "false", "accepting a belief", "believer", "advocating", "belief", "accepting"],
 "type": "metaphor"
}
  • source: A list of terms, which is the source of the relation mapping from.
  • target_random: A list of terms, where we want to find a mapping from source to.
  • target: A correctly ordered target_random that aligns with the source.

Given source and target_random, the task is to predict the correct order of target_random so that it matches target. In average 7 terms are in the set, so the total number of possible order is 5040.

Data Splits

name test
relation_mapping 45

Citation Information 

@article{czinczoll2022scientific,
  title={Scientific and Creative Analogies in Pretrained Language Models},
  author={Czinczoll, Tamara and Yannakoudakis, Helen and Mishra, Pushkar and Shutova, Ekaterina},
  journal={arXiv preprint arXiv:2211.15268},
  year={2022}
}