PoetryMTEB/ArabicPoetryThemeClustering

Arabic Poetry Theme Clustering

Clustering benchmark of Arabic poetry by theme/category labels for PoetryMTEB / MTEB-style embedding evaluation.

Derived from alwalid54321/Arabic_Poems (originally linked to arbml/ashaar al-diwan).

Dataset Card

Item	Description
Dataset version (PoetryMTEB)	1.0.0
Task	Clustering (unsupervised embedding evaluation)
Languages	Arabic (ar)
Unit	Full poem text (poem; verses joined by newlines)
Labels	18 themes (Arabic label_name + English/Chinese glosses)
Size	test=8875 (no train/validation — not required for MTEB Clustering)
Extra metadata	author, author_desc, era, meter
Evaluation metrics	MTEB Clustering: typically V-measure

Label taxonomy (18)

id	label_name (ar)	name_en	name_zh	theme_code	count
0	قصيدة قصيره	Short poem	短诗	T5	3438
1	قصيدة عامه	General / miscellaneous	一般/杂诗	T3	2507
2	قصيدة مدح	Praise / panegyric	颂赞	T4	728
3	قصيدة رومنسيه	Romantic	浪漫	T6	531
4	قصيدة حزينه	Sad / melancholic	悲伤	T10	428
5	قصيدة عتاب	Reproach / blame	责备	T13	270
6	قصيدة شوق	Longing / yearning	思慕	T7	200
7	قصيدة هجاء	Satire / invective	讽刺	T11	199
8	قصيدة غزل	Ghazal / love lyric	情诗	T9	147
9	قصيدة رثاء	Elegy	挽歌	T12	120
10	قصيدة دينية	Religious	宗教	T1	117
11	قصيدة فراق	Separation / farewell	离别	T14	89
12	قصيدة ذم	Censure / condemnation	贬责	T2	62
13	قصيدة وطنيه	Patriotic	爱国	T8	32
14	قصيدة اعتذار	Apology	致歉	T18	3
15	قصيدة المعلقات	Muʿallaqāt (hanging odes)	悬诗	T17	2
16	قصيدة الاناشيد	Anthems / songs	颂歌/歌曲	T16	1
17	قصيدة سياسية	Political	政治	T15	1

Codebook: label_taxonomy.json.

Features

Field	Type	Description
id	string	Example id (e.g. poem_00001)
title	string	Poem title
author	string	Poet name
author_desc	string	Poet biography / description
era	string	Historical era
meter	string	Prosodic meter (بحر …)
poem	string	Clustering input (parsed from legacy stringified verse list text)
label	int64	Cluster class id (frequency-ordered)
label_name	string	Arabic theme (formerly theme)
label_name_en	string	English gloss
label_name_zh	string	Chinese gloss
theme_code	string	Legacy code (T1…T18)
language	string	Always ar

Construction method

1. Load existing Hub test parquet.
2. Parse text (Python-list string of verses) → newline-joined poem.
3. Map themes to integer label + label_name_en / label_name_zh.
4. Keep test-only split; write Dataset Card + dataset_infos.yaml + taxonomy.

Supporting materials

Path	Content
VERSION / LICENSE / CITATION.* / metadata.json	Packaging metadata
label_taxonomy.json	AR / EN / ZH codebook
checksums.sha256	File checksums
scripts/evaluate_arabic_theme_clustering.py	Baseline V-measure probe

How to load

from datasets import load_dataset

ds = load_dataset("PoetryMTEB/ArabicPoetryThemeClustering")
ex = ds["test"][0]
print(ex["poem"][:200])
print(ex["label_name"], ex["label_name_en"], ex["label_name_zh"], ex["label"])

Intended use

PoetryMTEB / MTEB Clustering probing of Arabic poem embeddings by thematic category.

Citation / provenance

Please cite the upstream Arabic poetry resources used:

• alwalid54321/Arabic_Poems
• arbml/ashaar

This packaging: PoetryMTEB/ArabicPoetryThemeClustering (version 1.0.0)

License

Upstream license is not clearly stated on the intermediate packaging; treat as research use and follow the terms of alwalid54321/Arabic_Poems / arbml/ashaar.

---

阿拉伯诗歌主题聚类（中文说明）

用于 PoetryMTEB 聚类评测：仅 test 划分。每条样本提供阿拉伯语主题名及 英文 / 中文 对照（label_name_en / label_name_zh）。