README.md

---
license: gpl-3.0
task_categories:
- text-classification
task_ids:
- multi-class-classification
language:
- bn
multilinguality:
- monolingual
size_categories:
- 1K<n<10K
pretty_name: BengaliPoemThemeClassification
tags:
- poetry
- bangla
- bengali
- theme-classification
- genre-classification
- text-classification
- mteb
- poetrymteb
- embedding-evaluation
- stylometry
annotations_creators:
- expert-generated
source_datasets:
- shuhanmirza/Bengali-Poem-Dataset
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
  default: true
dataset_info:
- config_name: default
  features:
  - name: id
    dtype: string
  - name: title
    dtype: string
  - name: author
    dtype: string
  - name: poem
    dtype: string
  - name: label
    dtype: int64
  - name: label_name
    dtype: string
  splits:
  - name: train
    num_examples: 4134
  - name: validation
    num_examples: 518
  - name: test
    num_examples: 518
---

# BengaliPoemThemeClassification

Single-label **Bengali poem theme / subject classification** for PoetryMTEB, derived from the [Bengali Poem Dataset](https://github.com/shuhanmirza/Bengali-Poem-Dataset) (stylometric corpus).

Only **12 theme-oriented** `CLASS` labels are kept; formal genre labels (sonnet, nursery rhyme, epic, etc.) are excluded.

## Dataset Card

| Item | Description |
|------|-------------|
| **Source** | [shuhanmirza/Bengali-Poem-Dataset](https://github.com/shuhanmirza/Bengali-Poem-Dataset) |
| **Paper** | [A Stylometric Dataset for Bengali Poems](https://doi.org/10.1145/3582768.3582788) (NLPIR 2022 / ACM, 2023) |
| **Languages** | Bangla / Bengali (`bn`) |
| **Size** | **5170** poems (from 6070; dropped 900 non-theme / form-labeled); train=4134; validation=518; test=518 |
| **Classes** | 12 theme labels |
| **Splits** | Stratified by theme ≈ 80% / 10% / 10%, seed=42 |
| **License** | [GPL-3.0](https://www.gnu.org/licenses/gpl-3.0.html) (same as upstream repo) |
| **Evaluation metrics** | Classification on embeddings: **accuracy**, **macro/weighted F1** |

---

## Split statistics

| Split | #poems | ratio |
|-------|-------:|------:|
| train | 4134 | 79.96% |
| validation | 518 | 10.02% |
| test | 518 | 10.02% |
| **total** | **5170** | 100% |

### Per-theme counts by split

| id | label_name | romanization | gloss | train | validation | test | total |
|---:|------------|--------------|-------|------:|-----------:|-----:|------:|
| 0 | প্রেমমূলক | Premamūlaka | Love / romantic | 1084 | 136 | 136 | 1356 |
| 1 | চিন্তামূলক | Cintāmūlaka | Contemplative / philosophical | 1052 | 132 | 132 | 1316 |
| 2 | মানবতাবাদী | Mānabatābādī | Humanist | 579 | 72 | 72 | 723 |
| 3 | ভক্তিমূলক | Bhaktimūlaka | Devotional | 316 | 39 | 39 | 394 |
| 4 | রূপক | Rūpaka | Allegorical / metaphorical | 314 | 39 | 39 | 392 |
| 5 | প্রকৃতিমূলক | Prakṛtimūlaka | Nature | 277 | 35 | 35 | 347 |
| 6 | স্বদেশমূলক | Svadeśamūlaka | Patriotic | 199 | 25 | 25 | 249 |
| 7 | নীতিমূলক | Nītimūlaka | Didactic / moral | 144 | 18 | 18 | 180 |
| 8 | হাস্যরসাত্মক | Hāsyarasātmaka | Humorous | 101 | 13 | 13 | 127 |
| 9 | শোকমূলক | Śokamūlaka | Elegiac / sorrowful | 43 | 6 | 6 | 55 |
| 10 | ব্যঙ্গাত্মক | Byangātmaka | Satirical | 17 | 2 | 2 | 21 |
| 11 | স্তোত্রমূলক | Stotramūlaka | Hymn / praise | 8 | 1 | 1 | 10 |
| | **Sum** | | | **4134** | **518** | **518** | **5170** |

Excluded form-oriented labels: `সনেট`, `ছড়া`, `কাহিনীকাব্য`, `গীতিগাথা`, `মহাকাব্য`, `গাথাকাব্য`, `নাট্যগীতি`, `গীতিনাটিকা`, `লিপিমূলক`.

Codebook: `label_taxonomy.json`.

---

## Features

| Field | Type | Description |
|-------|------|-------------|
| `id` | string | Example id |
| `title` | string | Poem title (folder name) |
| `author` | string | Poet name (parent folder) |
| `poem` | string | Poem body |
| `label` | int64 | Theme class index |
| `label_name` | string | Theme label in Bangla (`CLASS.txt`) |

---

## Construction method

1. Scan [Bengali-Poem-Dataset](https://github.com/shuhanmirza/Bengali-Poem-Dataset) `dataset/{poet}/{title}/`.
2. Read `CLASS.txt`; keep only the 12 theme labels listed above.
3. Load poem `.txt` (excluding `CLASS.txt` / `SOURCE.txt`).
4. Stratified train / validation / test split by theme.

---

## How to load

```python
from datasets import load_dataset

ds = load_dataset("PoetryMTEB/BengaliPoemThemeClassification")
print(ds["test"][0]["label_name"], ds["test"][0]["title"])
```

---

## License

Distributed under **GNU General Public License v3.0 (GPL-3.0)**, consistent with the upstream GitHub repository.

---

## Citation / provenance

Please cite the original stylometric dataset paper and repository:

```bibtex
@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},
isbn = {9781450397629},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3582768.3582788},
doi = {10.1145/3582768.3582788},
booktitle = {Proceedings of the 2022 6th International Conference on Natural Language Processing and Information Retrieval},
pages = {176–180},
numpages = {5},
location = {Bangkok, Thailand},
series = {NLPIR '22}
}
```

- GitHub: [https://github.com/shuhanmirza/Bengali-Poem-Dataset](https://github.com/shuhanmirza/Bengali-Poem-Dataset)
- Paper: [https://doi.org/10.1145/3582768.3582788](https://doi.org/10.1145/3582768.3582788)
- This Hub packaging: `PoetryMTEB/BengaliPoemThemeClassification`