Harry214/paleography-web-text-triage-dataset

Paleography Web Text Triage Dataset

This dataset contains 400 manually labeled Chinese text snippets for an embedding-based text classification task in Ancient Chinese paleography corpus construction.

The task is to classify short web or reference snippets into four categories:

Label	Meaning
ksd	Scholarly or explanatory discussion about paleography, oracle bones, bronze inscriptions, excavated texts, or script history.
kpt	Primary transcription material such as oracle-bone, bronze inscription, or excavated-text snippets.
kde	Dictionary-style or reference-style entries about characters, words, pronunciations, forms, or glosses.
noise	Irrelevant, navigational, administrative, advertising, broken OCR, or otherwise unsuitable text.

Splits

Split	Rows
train	280
validation	60
test	60

Each split is balanced, with 70/15/15 examples per label.

Fields

• id: unique sample identifier
• text: Chinese input snippet
• label: one of ksd, kpt, kde, or noise
• source_url: public source URL or source reference
• language: currently chinese

Intended Use

This dataset is intended for small-scale experiments in corpus filtering, document routing, and information retrieval preparation for Ancient Chinese paleography materials.

Limitations

The dataset is small and domain-specific. All final samples are Chinese, so performance on English or mixed-language text is not evaluated. Some labels are inherently ambiguous because short scholarly comments may resemble dictionary entries or primary transcriptions.

Privacy and Ethics

The dataset uses public scholarly, reference, transcription, and webpage snippets. It does not intentionally include private or sensitive personal information.