Datasetpublic

yssf-io/service-public

Service-Public.fr — Fiches pratiques Administrative knowledge base extracted from service-public.fr, the French government's official public service information website, maintained by the Direction de l'information légale et administrative (DILA). Intended for use as context in RAG (Retrieval-Augmented Generation) pipelines for French administrative Q&A. Dataset details Total records: 5,535 Records with prose content (has_content = true): 3,055 Average text… See the full description on the dataset page: https://huggingface.co/datasets/yssf-io/service-public.

sourceHugging Faceetalab-2.0updated 7mo agoView on Hugging Face

0likes27downloads

Dataset Card

Service-Public.fr — Fiches pratiques

Administrative knowledge base extracted from service-public.fr, the French government's official public service information website, maintained by the Direction de l'information légale et administrative (DILA).

Intended for use as context in RAG (Retrieval-Augmented Generation) pipelines for French administrative Q&A.

Dataset details

—Total records: 5,535
—Records with prose content (has_content = true): 3,055
—Average text length (content records): ~9,652 characters
—Language: French
—Last source update: see last_modified field per record
—License: Licence Ouverte Etalab 2.0 (equivalent to CC-BY)

Fields

Field	Type	Description
`id`	string	Fiche identifier (e.g. `F1`, `F10004`)
`url`	string	Canonical URL on service-public.fr
`type`	string	Publication type (see table below)
`title`	string	Full title of the fiche
`description`	string	Short summary (1–2 sentences)
`subject`	string	Broad subject area
`theme`	string	Navigation theme
`audience`	string	Target audience (`Particuliers`, `Professionnels`, etc.)
`last_modified`	string	Date of last significant update (`YYYY-MM-DD`)
`legal_sources`	string	Comma-separated Légifrance URLs underpinning the fiche
`breadcrumb`	list[string]	Navigation path from homepage to this fiche
`text`	string	Full fiche content as clean Markdown
`char_count`	int	Character count of `text`
`has_content`	bool	`true` for fiches with prose content; `false` for forms, simulators, online services, etc.

Publication types

Type	Count
Fiche d'information conditionnée	1486
Fiche Question-réponse conditionnée	1404
Formulaire	882
Téléservice	700
Modèle de document	329
Dossier	232
Outil de recherche	177
Simulateur	171
Recherche guidée	96
Fiche Comment faire si conditionné	21
Fiche avec liens externes	15
Theme	11
	5
Téléservice personnalisé sur SP	2
Accueil audience	1
Liste - Comment faire si	1
Liste - Question-réponse	1
Liste - Service en ligne	1

Filtering tip

For RAG or QA use, filter to records where has_content = true to exclude forms, online services, and simulators that contain no prose:

python

from datasets import load_dataset

ds = load_dataset("yssf-io/service-public")
fiches = ds["train"].filter(lambda r: r["has_content"])

Text format

The text field uses Markdown conventions:

—## / ### headings for chapters and subchapters (titles are often phrased as questions)
—**bold** for emphasized terms
—> **À noter** / **Attention** / **À savoir** blockquotes for callouts and warnings
—Bullet lists for conditions, steps, and document requirements
—Conditional branches (e.g. different rules by company size or nationality) are preserved as nested headings

Source & license

Data sourced from the data.gouv.fr open data portal under the Licence Ouverte Etalab 2.0. © Direction de l'information légale et administrative (DILA) — Premier ministre.