PoetryMTEB/Appreciation-of-Chinese-Classical-Poetry

Appreciation of Chinese Classical Poetry

Chinese classical poetry with paired metadata and five-aspect literary analyses for PoetryMTEB / MTEB-style evaluation and computational poetics research.

Poems are drawn from expert appreciation volumes (mainly Shanghai Lexicographical Publishing House dictionaries). The released analysis fields are LLM distillations (DeepSeek-V3.1) of those expert appreciation texts into five free-text facets. The original long-form appreciation prose is not redistributed in this packaging.

corpus and analysis are two peer Hub configs joined by shared id (1:1 alignment; n=5574).

Dataset Card

Item	Description
Dataset version (PoetryMTEB)	1.1.1
Hub repo	PoetryMTEB/Appreciation-of-Chinese-Classical-Poetry
Language	Classical / literary Chinese poems (zh)
Unit	Full poem text (text) + five analysis strings
Configs	corpus (default) · analysis
Size	5,574 poems × 2 configs
Authors covered	1,454
Dynasty labels (raw)	28 distinct strings (not normalized)
Analysis model	DeepSeek-V3.1
License	CC BY-NC 4.0 (cc-by-nc-4.0)
Intended use	Embedding / retrieval probing; literary-analysis generation; educational NLP

Collections (configs)

Config	Hub split	n	Content
corpus (default)	test	5,574	Poem metadata + full text
analysis	test	5,574	Five free-text analysis facets linked by id

from datasets import load_dataset

corpus = load_dataset("PoetryMTEB/Appreciation-of-Chinese-Classical-Poetry", "corpus", split="test")
analysis = load_dataset("PoetryMTEB/Appreciation-of-Chinese-Classical-Poetry", "analysis", split="test")
assert set(corpus["id"]) == set(analysis["id"])

Features

corpus

Field	Type	Description
id	string	Stable poem id (shared with analysis)
dynasty	string	Dynasty / period label as recorded upstream (heterogeneous strings)
author	string	Author name
title	string	Title
type	string	Genre tag as released (currently all 诗 in this packaging)
text	string	Full poem text

analysis

Field	Type	Description (Chinese facet)
id	string	Join key → corpus.id
intent	string	创作意图 — creative / rhetorical intent
subject	string	题材 — subject matter / topical category
theme	string	主题 — thematic proposition
thought	string	思想 — intellectual / philosophical stance
emotion	string	情感 — affective tone / emotional expression

All five analysis fields are non-empty free-text summaries (character-level means ≈ intent 43.5, subject 14.3, theme 25.3, thought 39.6, emotion 38.7).

Construction overview

Expert appreciation dictionaries (12 volumes)
        │
        ├─► poem metadata + text ──────────────────────────────► config: corpus
        │
        └─► appreciation prose (not released)
                    │
                    ▼
              DeepSeek-V3.1 JSON distillation
              (intent / subject / theme / thought / emotion)
                    │
                    ▼
              config: analysis   (joined to corpus by id)

Distillation prompt (original)

Analysis fields were generated with DeepSeek-V3.1 using the following prompt:

prompt = f"""
            作为中国古典诗歌专家，请根据以下鉴赏文本分析该诗的创作意图、题材、主题、思想和情感：
            
            朝代：{dynasty}
            作者：{author}
            标题：《{title}》
            类型：{poem_type}
            
            【鉴赏文本】：
            {appreciation}
            
            请以严格JSON格式返回分析结果：
            {{
                "创作意图": "",
                "题材": "",
                "主题": "",
                "思想": "",
                "情感": ""
            }}
            
            要求：
            1. 确保只返回标准JSON，不要包含额外内容。"""

Packaging maps JSON keys 创作意图 / 题材 / 主题 / 思想 / 情感 to fields intent / subject / theme / thought / emotion.

Dynasty distribution (top labels)

Raw dynasty strings are not merged in this release (e.g. 清 vs 清代, 明 vs 明代 appear separately). Top counts:

dynasty (as released)	n
唐	1671
宋	1088
清代	419
明代	381
周	301
梁	253
清	248
近代	242
辽·金·元	149
晋	144
汉	92
魏	83
南朝	66
明	55
齐	54
(other labels)	328

Data sources

Compiled from expert appreciation publications (poem selection / commentary tradition). Publisher copyright on dictionary articles remains with the rights holders; this Hub package redistributes poem texts selected for PoetryMTEB research packaging plus LLM-derived facet summaries.

#	Title	Editors / authors	Publisher	Date	ISBN
1	先秦诗鉴赏辞典(新一版)	姜亮夫 夏传才 赵逵夫 郭维森 等	上海辞书出版社	2016-10	978-7-5326-4676-0
2	汉魏六朝诗鉴赏辞典(新一版)	袁行霈 周勋初 赵昌平 骆玉明 吴小如 王运熙 章培恒 曹道衡等	上海辞书出版社	2016-01	978-7-5326-4454-4
3	唐诗鉴赏辞典(新一版)	俞平伯 萧涤非 周汝昌 施蛰存 程千帆 周振甫 钱仲联 马茂元 霍松林等	上海辞书出版社	2013-08	978-7-5326-3981-6
4	宋诗鉴赏辞典(新一版)	葛晓音 陈伯海 赵昌平 莫砺锋 刘永翔 缪钺 霍松林 周振甫 吴调公 曾枣庄等	上海辞书出版社	2017-07	978-7-5326-4366-0
5	元明清诗鉴赏辞典(新一版)	钱仲联 章培恒 陈祥耀 潘啸龙等	上海辞书出版社	2018-03	978-7-5326-5055-2
6	历代女性诗词鉴赏辞典(新一版)	胡晓明主编	上海辞书出版社	2016-09	978-7-5326-4602-9
7	宋词鉴赏辞典(新一版)	夏承焘 唐圭璋 缪钺 叶嘉莹 周汝昌 宛敏灏 万云骏 钟振振等	上海辞书出版社	2013-08	978-7-5326-3982-3
8	唐诗选	中国社会科学院文学研究所选注	人民文学出版社	2009-04	978-7-0200-7243-9
9	唐诗三百首鉴赏辞典(文通版)	上海辞书出版社文学鉴赏辞典编纂中心编	上海辞书出版社	2017-04	978-7-5326-4914-3
10	白居易诗文鉴赏辞典	上海辞书出版社文学鉴赏辞典编纂中心编	上海辞书出版社	2014-08	978-7-5326-4221-2
11	元明清词鉴赏辞典(新一版)	钱仲联 马兴荣 叶嘉莹 王兆鹏 钟振振等	上海辞书出版社	2017-08	978-7-5326-4942-6
12	唐宋词鉴赏辞典(新一版)	唐圭璋 缪钺 叶嘉莹 周汝昌 俞平伯 施蛰存等	上海辞书出版社	2016-01	978-7-5326-4553-4

How to load / join

from datasets import load_dataset
import pandas as pd

corpus = load_dataset("PoetryMTEB/Appreciation-of-Chinese-Classical-Poetry", "corpus", split="test")
analysis = load_dataset("PoetryMTEB/Appreciation-of-Chinese-Classical-Poetry", "analysis", split="test")

df = pd.merge(
    corpus.to_pandas(),
    analysis.to_pandas(),
    on="id",
    how="inner",
)
print(len(df), df.columns.tolist())
print(df.iloc[0][["title", "author", "theme", "emotion"]])

Supporting materials

Path	Content
VERSION / LICENSE / CITATION.* / metadata.json	Packaging metadata
dataset_infos.yaml	Hub viewer configs + sizes
checksums.sha256	File hashes
corpus/*.parquet / analysis/*.parquet	Data files

Intended use

• PoetryMTEB monolingual Chinese classical-poetry probing (retrieval / clustering / generation helpers).
• Training or evaluating models that predict literary facets from poem text.
• Educational demos of computational appreciation (with human oversight).

Limitations

• Analysis labels are machine-generated distillations, not human gold annotations of the five facets.
• Source appreciation prose is omitted; users cannot recover the original dictionary article from this release.
• dynasty / type fields inherit upstream string noise and are not a curated taxonomy.
• Classical poem texts may still be subject to edition-level rights constraints depending on jurisdiction; use under CC BY-NC 4.0 for this packaging layer and respect upstream dictionary copyrights.

Citation

@misc{appreciation_of_chinese_classical_poetry,
  title     = {Appreciation of Chinese Classical Poetry},
  author    = {PoetryMTEB Contributors},
  year      = {2026},
  version   = {1.1.1},
  url       = {https://huggingface.co/datasets/PoetryMTEB/Appreciation-of-Chinese-Classical-Poetry},
  note      = {PoetryMTEB packaging; poem corpus + DeepSeek-V3.1 facet distillation from expert appreciation dictionaries}
}

Please also cite the relevant source dictionaries listed above when redistributing research that depends on their selection / commentary tradition.

License

Distributed under CC BY-NC 4.0 (Creative Commons Attribution–NonCommercial 4.0 International). See LICENSE.

Non-commercial research and embedding-evaluation use with attribution. This packaging does not grant rights to reproduce the full commercial dictionary articles.

Hub packaging: PoetryMTEB/Appreciation-of-Chinese-Classical-Poetry (version 1.1.1); parquet sizes corpus=1,414,281 B, analysis=1,772,469 B.