PoetryMTEB/BengaliPoetryThemeClustering

Bengali Poetry Theme Clustering

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

Related sources: rishiraj/bengali-poems, Bengali-Poem-Dataset (ACM SIGIR 2023 resource paper, DOI).

Dataset Card

Item	Description
Dataset version (PoetryMTEB)	1.0.0
Task	Clustering (unsupervised embedding evaluation)
Languages	Bengali (bn)
Unit	Full poem text (poem)
Labels	21 category tags (Bengali label_name + English/Chinese glosses)
Size	test=6080 (no train/validation — not required for MTEB Clustering)
License	GPL-3.0 (aligned with upstream Bengali Poem Dataset packaging)
Evaluation metrics	MTEB Clustering: typically V-measure (K-Means on embeddings)

Label note: Upstream category mixes theme tags (e.g. love, nature) and form/genre tags (e.g. sonnet, epic). See label_kind. Companion classification release with theme-only filtering: PoetryMTEB/BengaliPoemThemeClassification.

Label taxonomy (21)

id	label_name (bn)	name_en	name_zh	kind	theme_code	count
0	প্রেমমূলক	Love / romantic	爱情	theme	T4	1356
1	চিন্তামূলক	Contemplative / philosophical	哲思	theme	T5	1320
2	মানবতাবাদী	Humanist	人文主义	theme	T3	723
3	সনেট	Sonnet	十四行诗	form	T11	468
4	রূপক	Allegorical / metaphorical	讽喻	theme	T10	395
5	ভক্তিমূলক	Devotional	虔敬	theme	T1	394
6	প্রকৃতিমূলক	Nature	自然	theme	T8	347
7	ছড়া	Nursery rhyme / children's verse	儿歌韵文	form	T9	326
8	স্বদেশমূলক	Patriotic	爱国	theme	T6	249
9	নীতিমূলক	Didactic / moral	训诫	theme	T7	180
10	হাস্যরসাত্মক	Humorous	幽默	theme	T16	127
11	কাহিনীকাব্য	Narrative poetry	叙事诗	form	T14	57
12	শোকমূলক	Elegiac / sorrowful	哀挽	theme	T2	55
13	গীতিগাথা	Lyrical ballad	抒情歌谣	form	T15	31
14	ব্যঙ্গাত্মক	Satirical	讽刺	theme	T12	21
15	স্তোত্রমূলক	Hymn / praise	颂歌	theme	T17	10
16	মহাকাব্য	Epic	史诗	form	T18	9
17	গাথাকাব্য	Ballad / narrative verse	歌谣体	form	T19	4
18	নাট্যগীতি	Dramatic lyric	戏剧抒情	form	T20	3
19	লিপিমূলক	Epistolary	书信体	form	T13	3
20	গীতিনাটিকা	Musical / lyrical play	抒情小剧	form	T21	2

• Theme-like labels: 5177 poems · Form/genre labels: 903 poems
• Codebook: label_taxonomy.json

Features

Field	Type	Description
id	string	Example id (e.g. poem_00001)
title	string	Poem title (empty if unknown)
author	string	Poet name
poem	string	Clustering input (formerly text)
label	int64	Cluster class id (0 … 20, frequency-ordered)
label_name	string	Bengali category name (formerly theme)
label_name_en	string	English gloss
label_name_zh	string	Chinese gloss
theme_code	string	Legacy code (T1…T21) from prior packaging
label_kind	string	theme or form
language	string	Always bn

Construction method

1. Load existing PoetryMTEB/BengaliPoetryThemeClustering test parquet.
2. Map Bengali categories to integer label + label_name_en / label_name_zh.
3. Rename text→poem, theme→label_name; clear placeholder titles (未命名诗歌 …).
4. Keep test-only split (MTEB Clustering protocol).
5. Write Dataset Card, dataset_infos.yaml, taxonomy, checksums, eval helper.

Supporting materials

Path	Content
VERSION	Packaging version
LICENSE	GPL-3.0 notice
CITATION.cff / CITATION.bib	Citation metadata
metadata.json	Dataset metadata
label_taxonomy.json	Full BN / EN / ZH codebook
checksums.sha256	File checksums
scripts/evaluate_bengali_theme_clustering.py	Baseline V-measure probe

How to load

from datasets import load_dataset

ds = load_dataset("PoetryMTEB/BengaliPoetryThemeClustering")
ex = ds["test"][0]
print(ex["poem"][:120])
print(ex["label_name"], ex["label_name_en"], ex["label_name_zh"], ex["label"])

Intended use

• PoetryMTEB / MTEB Clustering probing of Bengali poem embeddings.
• Not a supervised classification split (use BengaliPoemThemeClassification for train/val/test classification).

Citation

@inproceedings{10.1145/3582768.3582788,
  author    = {Shuhan, Mirza Kamrul Bashar and Dey, Rupasree and Saha, Sourav and Anjum, Md Shafa Ul and Zaman, Tarannum Shaila},
  title     = {A Stylometric Dataset for Bengali Poems},
  year      = {2023},
  booktitle = {Proceedings of the 2022 6th International Conference on Natural Language Processing and Information Retrieval},
  pages     = {176--180},
  publisher = {ACM},
  doi       = {10.1145/3582768.3582788},
  url       = {https://doi.org/10.1145/3582768.3582788},
  series    = {NLPIR '22}
}

Upstream: https://github.com/shuhanmirza/Bengali-Poem-Dataset · rishiraj/bengali-poems
This packaging: PoetryMTEB/BengaliPoetryThemeClustering (version 1.0.0)

License

GPL-3.0. See upstream Bengali Poem Dataset terms.

---

孟加拉诗歌主题聚类（中文说明）

用于 PoetryMTEB 聚类评测的孟加拉语诗歌数据集：仅 test 划分（MTEB Clustering 无需 train/val）。

每条样本提供孟加拉语类别名 label_name，以及 英文 label_name_en、中文 label_name_zh。上游类别混有主题与体裁标签，见 label_kind。