English | فارسی

Ganjoor-Corpus

The Ganjoor poetry corpus as four tables covering poems, books of poetry, and poet information. This corpus can be used for training models, statistical work, and building other datasets.

Configs

poets: 234 poets.

field	description
poet_id	Ganjoor poet id
name, nickname	full name and pen name
url	Ganjoor path
birth_year, death_year	lunar Hijri; birth_year_valid / death_year_valid say whether Ganjoor marks the date as reliable
birth_place, death_place	as given by Ganjoor
description	poet biography

categories: where a poem sits in the Ganjoor category tree, including book name, section within the book, and the rest.

field	description
cat_id	Ganjoor category id
parent_id	parent category; null at the poet root
poet_id	owning poet
title, url	category title and path
book_name	filled when Ganjoor sets it on the poet root (e.g. دیوان حافظ)
description	category note, if any
depth	0 poet, 1 work (گلستان, شاهنامه, غزلیات), deeper = chapter

poems (default): 132,591 poems.

field	description
poem_id	Ganjoor poem id
poet_id	Ganjoor poet id
cat_id	leaf folder
book_id	work-level category (depth 0 or 1)
title, full_title, url	poem title and path
rhythm, metre_id	aruz string and metre id, empty when Ganjoor has none
rhyme	RhymeLetters
poem_format	form from poem sections (e.g. Ghazal, Masnavi); `
language	Ganjoor language tag when present
source	text source name on Ganjoor
summary, summary_is_ai	poem summary; summary_is_ai is true when it starts with «هوش مصنوعی»
couplet_count	number of rows this poem contributes to couplets

HTML is not included. 127,695 of 129,414 summaries are AI-generated.

couplets: 1,488,130 beyts. One row per couplet, each mesra in its own column.

field	description
first, second	the two mesras of the couplet
poem_id, poet_id	join keys
couplet_index	couplet number within the poem (Ganjoor CoupletIndex)

Right / CenteredVerse1 → first; Left / CenteredVerse2 → second. Paragraphs and comments are omitted (80,555 prose verses). An unpaired mesra is kept with an empty second (4,859 rows).

from datasets import load_dataset

poems = load_dataset("farbodbij/Ganjoor-Corpus", "poems", split="train")
couplets = load_dataset("farbodbij/Ganjoor-Corpus", "couplets", split="train")
poets = load_dataset("farbodbij/Ganjoor-Corpus", "poets", split="train")
categories = load_dataset("farbodbij/Ganjoor-Corpus", "categories", split="train")

How it was built

Fields in this dataset (metre, rhyme, format, summaries, biographies, and the rest) are taken from Ganjoor as-is, with no correction or rewriting. Their accuracy therefore depends on the accuracy of Ganjoor’s data.

Thanks Disclaimer

We are thankful to Ganjoor team for releasing their meticulously curated data publicly, allowing anyone to deliver high-quality Persian datasets to the NLP community. This both helps Persian as a low-resource language and also contributes to distribution of the Iranian culture.