---
license: unknown
task_categories:
- other
language:
- en
multilinguality:
- monolingual
size_categories:
- 10K<n<100K
pretty_name: English Poetry Topic Clustering
tags:
- poetry
- english
- topic-clustering
- clustering
- mteb
- poetrymteb
- embedding-evaluation
annotations_creators:
- found
source_datasets:
- AJMC2002/poems
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
  default: true
dataset_info:
- config_name: default
  features:
  - name: id
    dtype: string
  - name: author
    dtype: string
  - name: title
    dtype: string
  - name: poem
    dtype: string
  - name: label
    dtype: int64
  - name: label_name
    dtype: string
  - name: label_name_en
    dtype: string
  - name: label_name_zh
    dtype: string
  - name: language
    dtype: string
  splits:
  - name: test
    num_examples: 14334
---

# English Poetry Topic Clustering

**Clustering** benchmark of **English poetry** by topic keywords for PoetryMTEB / MTEB-style embedding evaluation.

Derived from [AJMC2002/poems](https://huggingface.co/datasets/AJMC2002/poems) (topic-tagged poems originally parsed from a Kaggle poetry collection).

## Dataset Card

| Item | Description |
|------|-------------|
| **Dataset version (PoetryMTEB)** | `1.0.0` |
| **Task** | Clustering (unsupervised embedding evaluation) |
| **Languages** | English (`en`) |
| **Unit** | Full poem text (`poem`) |
| **Labels** | **144** topic keywords (`label_name` = English; `label_name_zh` = Chinese) |
| **Size** | test=**14334** (**no train/validation** — not required for MTEB Clustering) |
| **Evaluation metrics** | Typically **V-measure** (K-Means on embeddings) |

> Prior Hub packaging exposed train/val/test (~6:2:2). This release **merges all poems into `test`** for standard MTEB Clustering probing.

## Label taxonomy (144 topics)

Preview (top by frequency; full list in `label_taxonomy.json`):

| id | label_name (en) | name_zh | count |
|---:|-----------------|---------|------:|
| 0 | `alone` | 孤独 | 100 |
| 1 | `america` | 美国 | 100 |
| 2 | `angel` | 天使 | 100 |
| 3 | `animal` | 动物 | 100 |
| 4 | `baby` | 婴儿 | 100 |
| 5 | `beach` | 海滩 | 100 |
| 6 | `beautiful` | 美丽 | 100 |
| 7 | `beauty` | 美 | 100 |
| 8 | `birth` | 出生 | 100 |
| 9 | `brother` | 兄弟 | 100 |
| 10 | `butterfly` | 蝴蝶 | 100 |
| 11 | `children` | 儿童 | 100 |
| 12 | `christmas` | 圣诞节 | 100 |
| 13 | `cinderella` | 灰姑娘 | 100 |
| 14 | `crazy` | 疯狂 | 100 |
| 15 | `dance` | 舞蹈 | 100 |
| 16 | `dark` | 黑暗 | 100 |
| 17 | `death` | 死亡 | 100 |
| 18 | `destiny` | 命运 | 100 |
| 19 | `dream` | 梦想 | 100 |
| … | *(all 144 topics)* | … | … |

## Features

| Field | Type | Description |
|-------|------|-------------|
| `id` | string | Upstream poem id |
| `author` | string | Poet name |
| `title` | string | Poem title |
| `poem` | string | **Clustering input** (formerly `text`) |
| `label` | int64 | Topic class id (0 … 143) |
| `label_name` | string | Topic keyword (formerly `topic`) |
| `label_name_en` | string | Same as `label_name` (English) |
| `label_name_zh` | string | Chinese gloss of the topic keyword |
| `language` | string | Always `en` |

## Construction method

1. Load prior `PoetryMTEB/EnglishPoetryTopicClustering` train / val / test.
2. Merge all rows into a single **test** split.
3. Map topics → integer `label` + Chinese glosses; rename `text`→`poem`.
4. Write parquet, Dataset Card, taxonomy, checksums, eval script.

## Supporting materials

| Path | Content |
|------|---------|
| `VERSION` / `LICENSE` / `CITATION.*` / `metadata.json` | Packaging metadata |
| `label_taxonomy.json` | Full EN/ZH topic codebook |
| `checksums.sha256` | File checksums |
| `scripts/evaluate_english_topic_clustering.py` | Baseline V-measure probe |

## How to load

```python
from datasets import load_dataset

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

## Citation / provenance

- Upstream: [AJMC2002/poems](https://huggingface.co/datasets/AJMC2002/poems)
- This packaging: `PoetryMTEB/EnglishPoetryTopicClustering` (version 1.0.0)

## License

Upstream license is **not clearly stated** (`AJMC2002/poems` / Kaggle provenance). Marked **unknown**; use for research and respect original poem copyrights.

---

# 英语诗歌主题聚类（中文说明）

MTEB **聚类**评测集：仅 **test=14334**，**144** 个英文话题关键词，并提供中文译名 `label_name_zh`。