Dataset Viewer
Auto-converted to Parquet Duplicate
head
stringlengths
1
18
tail
stringlengths
1
39
relation
stringclasses
22 values
cloak
carry
event
bottle
contain
event
pistol
bullet
mero
phone
likely
random
mackerel
retire
random
penguin
inhabit
event
spoon
libertarianism
random
peach
lemon
coord
coyote
bear
coord
hawk
spot
event
bookcase
date
random
revolver
produce
event
hotel
befriend
random
parsley
trekking
random
sieve
mine
random
alligator
likely
random
hotel
sit
event
carrot
edict
random
coyote
cat
coord
butterfly
cricket
coord
hawk
other
random
bottle
gbp
random
scooter
priority
random
hammer
authorisation
random
rifle
elector
random
coyote
leg
mero
cathedral
famous
attri
stove
Serb
random
sieve
disciple
random
knife
sharpen
event
blouse
hometown
random
violin
expensive
attri
coyote
rat
coord
train
move
event
glider
cockpit
mero
restaurant
eat
event
broccoli
radish
coord
rake
plier
coord
flute
saxophone
coord
clarinet
do
random
apricot
skin
mero
freezer
agenda
random
cat
donkey
coord
tiger
establish
random
cathedral
temple
hyper
hammer
spade
coord
whale
catfish
coord
spoon
side-by-side
random
washer
plug
mero
whale
elephant
coord
spear
clean
attri
car
board
random
falcon
see
event
falcon
redo
random
grape
endorsement
random
cat
breathe
event
oven
emulator
random
cockroach
ugly
attri
sweater
foster
random
cockroach
vols
random
rat
tatter
random
potato
wash
event
sheep
drink
event
cat
meticulous
random
beetle
brake
random
cello
banjo
coord
beetle
die
event
pub
block
random
cottage
cook
event
vest
instrumental
random
grenade
suitable
random
cranberry
fresh
attri
saw
baroness
random
axe
artifact
hyper
scarf
button
event
glove
pullover
coord
wasp
breathe
event
library
push
random
elephant
testimony
random
robin
handicap
random
knife
napkin
coord
squirrel
fast
attri
piano
large
attri
fork
wash
event
rifle
auditorium
random
cathedral
practice
random
goldfish
prospective
random
tanker
nerve
random
cypress
easy
random
sieve
create
random
couch
furniture
hyper
coat
heavy
attri
beet
cabbage
coord
glove
elegant
attri
motorcycle
sauce
random
bear
frighten
event
cathedral
bench
mero
fridge
white
attri
jacket
quote
random
cloak
anarchist
random
End of preview. Expand in Data Studio

relbert/lexical_relation_classification(複製)

データの内容

語彙関係分類のためのデータ。SphereRE で使われた 5 つのデータセット(BLESS、CogALexV、EVALution、K&H+N、ROOT09)を、それぞれ 1 つの設定(config)として収める。1 行が head と tail の語の組と、その関係 relation からなる(例: turtle と live が event)。すべての資源のライセンスは CC-BY-NC-4.0 で、学術や個人の研究には使えるが商用には制限がある。

データの概要

項目 値
行数 111,509(5 構成・14 分割の合計)
列数 3
容量 6.57 MB(ファイル 17 本)
ファイルの形式 jsonl 14、py 1
言語 en
公開 public

構成と分割

構成(config) 分割(split)ごとの行数 列数
BLESS train 18,582、test 6,637、validation 1,327 3
CogALexV train 3,054、test 4,260 3
EVALution train 5,160、test 1,846、validation 372 3
K&H+N train 40,256、test 14,377、validation 2,876 3
ROOT09 train 8,933、test 3,191、validation 638 3

列(構成 BLESS・分割 train。どの構成も同じ列)

列 型 先頭の行の値
head string cloak
tail string carry
relation string event

行数と列は HF のデータセットビューアの集計(元のデータセットの現在の版で数えたもの)。全データセットの一覧は HYUKI5/dataset-catalog にある。

複製について

relbert/lexical_relation_classification をリビジョン 9862d1e870fe6dba4922d3d326c9c8b90a2ecad5 の時点でそのまま複製したもの。中身は変えていない。判断モデル Laya(convaiinnovations/laya)の微調整に使う候補として、調査メモ(Laya 学習用データセット調査)に載せたデータを集めている。

出典

論文

アノテーションの方法

元のカードにアノテーションの節が無かった。下の「調査メモの記述」と「元のカード」を参照。

微調整に使うときの注意

  • 調査メモの台帳での区分: C(NC・ND・研究専用・評価専用など、商用の利用を妨げうる記載がある)。区分は記載の文言で機械的に付けたもので、法的な判定ではない。学習に入れる前に元のカードとライセンスの原文で確かめる。
  • Laya の入力は 2,048 token まで(ModernBERT-large / mmBERT)。長い本文は先頭から黙って切られるので、切れる行は事前に除くか、要約・切り出しをしてから入れる。
  • 分割は記事・文書単位で行い、同じ本文から作った別の問いを学習側と較正側に分けない。温度と閾値は学習に使っていない分割で別々に当てる。

調査メモの記述

メモ 節 ライセンス(記載) 確認 注記
ontology_kg_entity.md Q5. KG トリプルの真偽判定・常識知識・上位下位・時間関係・オントロジー整合のデータにどんなものがあり、Laya に載るか (L160) other 確認済 確認印なし(各ファイルの凡例により確認済扱い) / 原文注記: 語レベルの対で文脈なし。5 つのデータの原ライセンスが混在(other)。SemEval-2016/2018 の本体は未調査
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) other, CC BY-NC 4.0 (DPI) / license_use=non-commercial 確認済 [Jev 元データ source] source_annotation_converted(deterministic recast; no model r
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) other, CC BY-NC 4.0 (DPI) / license_use=non-commercial 確認済 [Jev 元データ source] source_annotation_converted(deterministic recast; no model r
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) other, CC BY-NC 4.0 (DPI) / license_use=non-commercial 確認済 [Jev 元データ source] source_annotation_converted(deterministic recast; no model r
jev_sources_tasksource_bekko.csv CSV(tasksource-jev-typed-decisions) other, CC BY-NC 4.0 (DPI) / license_use=non-commercial 確認済 [Jev 元データ source] source_annotation_converted(deterministic recast; no model r

元のカード(日本語訳)

以下は元のデータセットのカードの本文を日本語に訳したもの(機械翻訳)。訳に疑問があるときは、末尾の原文を確かめる。


"relbert/lexical_relation_classification" のデータセットカード

データセットの説明

データセットの概要

SphereRE で使われた、語彙関係分類のための 5 つのデータセット(BLESS、CogALexV、EVALution、K&H+N、ROOT09)。

データセットの概要

このデータセットには、Analogy Language Model で使われた 5 種類の単語の類推問題が含まれる。

name train validation test
BLESS 18582 1327 6637
CogALexV 3054 - 4260
EVALution 5160 372 1846
K&H+N 40256 2876 14377
ROOT09 8933 638 3191

データセットの構造

データの例

例は次のとおり。

{"head": "turtle", "tail": "live", "relation": "event"}

stem と tail が単語のペアで、relation が対応する関係のラベルである。

引用情報

@inproceedings{wang-etal-2019-spherere,
    title = "{S}phere{RE}: Distinguishing Lexical Relations with Hyperspherical Relation Embeddings",
    author = "Wang, Chengyu  and
      He, Xiaofeng  and
      Zhou, Aoying",
    booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/P19-1169",
    doi = "10.18653/v1/P19-1169",
    pages = "1727--1737",
    abstract = "Lexical relations describe how meanings of terms relate to each other. Typical examples include hypernymy, synonymy, meronymy, etc. Automatic distinction of lexical relations is vital for NLP applications, and also challenging due to the lack of contextual signals to discriminate between such relations. In this work, we present a neural representation learning model to distinguish lexical relations among term pairs based on Hyperspherical Relation Embeddings (SphereRE). Rather than learning embeddings for individual terms, the model learns representations of relation triples by mapping them to the hyperspherical embedding space, where relation triples of different lexical relations are well separated. Experiments over several benchmarks confirm SphereRE outperforms state-of-the-arts.",
}

ライセンス

すべての資源のライセンスは CC-BY-NC-4.0 に従う。したがって、学術目的または個人の研究には自由に利用できるが、商用利用には制限がある。

原文(英語)

Dataset Card for "relbert/lexical_relation_classification"

Dataset Summary

Five different datasets (BLESS, CogALexV, EVALution, K&H+N, ROOT09) for lexical relation classification used in SphereRE.

Dataset Summary

This dataset contains 5 different word analogy questions used in Analogy Language Model.

name train validation test
BLESS 18582 1327 6637
CogALexV 3054 - 4260
EVALution 5160 372 1846
K&H+N 40256 2876 14377
ROOT09 8933 638 3191

Dataset Structure

Data Instances

An example looks as follows.

{"head": "turtle", "tail": "live", "relation": "event"}

The stem and tail are the word pair and relation is the corresponding relation label.

Citation Information

@inproceedings{wang-etal-2019-spherere,
    title = "{S}phere{RE}: Distinguishing Lexical Relations with Hyperspherical Relation Embeddings",
    author = "Wang, Chengyu  and
      He, Xiaofeng  and
      Zhou, Aoying",
    booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
    month = jul,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/P19-1169",
    doi = "10.18653/v1/P19-1169",
    pages = "1727--1737",
    abstract = "Lexical relations describe how meanings of terms relate to each other. Typical examples include hypernymy, synonymy, meronymy, etc. Automatic distinction of lexical relations is vital for NLP applications, and also challenging due to the lack of contextual signals to discriminate between such relations. In this work, we present a neural representation learning model to distinguish lexical relations among term pairs based on Hyperspherical Relation Embeddings (SphereRE). Rather than learning embeddings for individual terms, the model learns representations of relation triples by mapping them to the hyperspherical embedding space, where relation triples of different lexical relations are well separated. Experiments over several benchmarks confirm SphereRE outperforms state-of-the-arts.",
}

LICENSE

The LICENSE of all the resources are under CC-BY-NC-4.0. Thus, they are freely available for academic purpose or individual research, but restricted for commercial use.

Downloads last month
30