Datasetpublic

Aananda-giri/gorkhapatra-nepali-epaper

Gorkhapatra Nepali E-Paper Corpus Per-article text extracted from PDF e-papers published on epaper.gorkhapatraonline.com, covering 11 newspaper slugs (gorkhapatra, risingnepal, friday-suppliment, madhuparka, muna, nayanepal, loksewa, saturday, yuwamunch, gorkhapatra-125, other). Extraction is layout-aware (geometry + font size, no ML model/fixed template) and reconstructs article boundaries — headline, dateline, and paragraphs in reading order — directly from the PDF's… See the full description on the dataset page: https://huggingface.co/datasets/Aananda-giri/gorkhapatra-nepali-epaper.

sourceHugging Faceotherupdated 2mo agoView on Hugging Face

0likes52downloads

Dataset Card

Gorkhapatra Nepali E-Paper Corpus

Per-article text extracted from PDF e-papers published on epaper.gorkhapatraonline.com, covering 11 newspaper slugs (gorkhapatra, risingnepal, friday-suppliment, madhuparka, muna, nayanepal, loksewa, saturday, yuwamunch, gorkhapatra-125, other).

Extraction is layout-aware (geometry + font size, no ML model/fixed template) and reconstructs article boundaries — headline, dateline, and paragraphs in reading order — directly from the PDF's glyph-level layout, converting legacy Preeti-family Nepali fonts to Unicode Devanagari as it goes.

—243,199 articles
—4,529 source PDFs
—Most content is Nepali (Devanagari script); friday-suppliment and risingnepal are in English.

Articles that jump across pages (a story ending "बाँकी पृष्ठ N मा" / continued on page N) are resolved before shipping: the continuation's paragraphs are spliced onto the source article and the continuation is dropped as a separate row, so text is always the complete article, never cut off mid-sentence. is_merged marks rows built this way. The rare case where a jump was declared but its continuation couldn't be matched (source stays truncated) is tracked separately for manual review, not shipped silently — see dangling_continuations.jsonl in the source repo.

Fields

field	type	description
`source_pdf`	string	original PDF filename
`slug`	string	newspaper slug (parent directory name)
`date`	string	BS (Bikram Sambat) date parsed from the filename, if present
`page`	int	page number the article starts on
`lang`	string	detected language of the article
`headline`	string	article headline
`dateline`	string	dateline, if present
`text`	string	full article text (paragraphs joined) — the field to train on
`paragraphs`	list[string]	paragraphs in reading order
`n_paragraphs`	int	paragraph count
`n_chars`	int	character count of `text`
`is_merged`	bool	true if this article's text was stitched together from a cross-page continuation
`price`	string	cover price printed on the issue's masthead, if parsed (e.g. "दस रुपियाँ")
`volume`	int	masthead volume number (वर्ष), if parsed
`issue`	int	masthead issue number (अङ्क), if parsed
`source_url`	string	original download URL on epaper.gorkhapatraonline.com

date/price/volume/issue are parsed from the issue's nameplate (masthead) and are only ~59% populated — the nameplate format is specific to Gorkhapatra-family layouts, so slugs like risingnepal/nayanepal/etc. don't carry it in parseable form. source_url is 100% populated.

Loading

python

from datasets import load_dataset

ds = load_dataset("aananda-giri/gorkhapatra-nepali-epaper")