Datasets:
GiantMelon
中文机器生成文本检测语料，外加一部分与中文维基配对的英文条目译文，以及 17 篇演示案例。这里描述的是本次上传的文件，不是论文表格的逐字复刻。
Document-level Chinese machine-generated-text detection data, plus English renderings of paired Chinese Wikipedia articles, plus 17 demonstration cases. Counts below are measured on this release.
There is no train/test split in the files. The full split is the whole release. A paper may describe a document-level 9:1 protocol; the seed is not stored here, so this repository does not invent one.
Subsets
Filter with the subset column.
| subset | rows | what it is |
|---|---|---|
news |
246,748 | Cleaned Chinese news. label=0: 123,889 human documents, 103.79M CJK characters. label=1: 122,859 AI documents, 243.60M CJK characters, generator=grok-4.1-fast. 122,859 ids occur on both sides (pair_id set). 1,030 human documents have no AI twin (pair_id empty). Punctuation already removed. |
wiki_claude |
3,964 | 1,982 cleaned Chinese pairs from the review subset. Human: 8.12M CJK characters, label=0. AI: 2.88M CJK characters, label=1, generator=claude-4.5-opus. pair_id is the Wikipedia id. |
wiki_parallel |
103,710 | 51,855 Chinese Wikipedia articles (20.59M CJK characters, punctuation kept) and 51,855 English renderings of the same ids. label is null: these rows are not human/AI detection labels. They are not the GPT-4o RAID parallel corpus. 75,221 additional English files had no Chinese counterpart and are not included. |
examples |
17 | Demonstration passages, not training data. 9 AI rows (doubao or deepseek) and 8 human rows. source names the origin. paper_p_ai is the score recorded with the case list. |
News CJK totals: human 103,787,963; AI 243,603,638. Wiki-Claude CJK totals: human 8,115,726; AI 2,877,254.
Columns
| column | meaning |
|---|---|
doc_id |
File stem, or the case id for examples |
text |
Document text |
lang |
zh or en |
subset |
news, wiki_claude, wiki_parallel, examples |
label |
0 human, 1 AI. Null for wiki_parallel |
generator |
Empty, grok-4.1-fast, claude-4.5-opus, doubao, or deepseek |
pair_id |
Shared id for a pair. Empty when a news document has no twin |
source |
Set on examples only. Empty elsewhere |
paper_p_ai |
Set on examples only. Null elsewhere |
Examples in this release
| doc_id | label | generator | source |
|---|---|---|---|
| 1-1 | 1 | doubao | Doubao / news |
| 1-2 | 1 | doubao | Doubao / encyclopedia |
| 1-3 | 1 | doubao | Doubao / literature |
| 2-1 | 1 | deepseek | Deepseek / news |
| 2-2 | 1 | deepseek | Deepseek / encyclopedia |
| 2-3 | 1 | deepseek | Deepseek / literature |
| 3-1 | 1 | doubao | Doubao / news (misclassified) |
| 3-2 | 1 | doubao | Doubao / encyclopedia (misclassified) |
| 3-3 | 1 | doubao | Doubao / literature (misclassified) |
| 4-1 | 0 | none | 朱自清《荷塘月色》 |
| 4-2 | 0 | none | 南方周末 2012 新年献词 |
| 4-3 | 0 | none | 王小波《白银时代》 |
| 5-1 | 0 | none | 左拉《萌芽》译著 |
| 5-2 | 0 | none | 塞万提斯《堂吉诃德》译著 |
| 6-1 | 0 | none | 艾青《大堰河──我的褓姆》 |
| 7-1 | 0 | none | 高光谱成像技术在生物医学中的应用进展 |
| 7-2 | 0 | none | 中新社 绥芬河 |
Literary works, translations, news articles, and the paper excerpt in examples remain under their original copyright. See LICENSE.
Loading
from datasets import load_dataset
ds = load_dataset("layce/giantmelon-zh", split="full")
news = ds.filter(lambda row: row["subset"] == "news")
The repository is gated. Agree to the terms on the dataset page before downloading.
Not included
- 75,221 English encyclopedia renderings that have no Chinese file with the same id
- Model weights
- The English word dictionary used by the English detector
- Downloads last month
- 17