oliverkinch/dst-table-prompts

DST Tables

News articles from Danmarks Statistik (Statistics Denmark) paired with the statistical tables they contain and the underlying raw data from StatBank.

Dataset Description

Each row is a "Nyt fra Danmarks Statistik" article that contains at least one inline statistical table. The dataset provides:

• full_article — the complete published article: prose text followed by the statistical tables as markdown

• inline_tables_md — the tables alone as markdown (useful as input for generation tasks)

• tables — the underlying raw data fetched from the StatBank API

• Source: dst.dk/nyt + api.statbank.dk

• Language: Danish

• License: CC BY 4.0

• Articles: 3,382

Dataset Structure

Column	Type	Description
article_id	string	Sequential article identifier
url	string	URL of the DST news article
content_type	string	Publication type (nyt)
title	string	Article title
date	string	Publication date in ISO format (e.g. 2026-04-22)
date_raw	string	Publication date as printed in Danish (e.g. 22. april 2026)
series	string	Article series name, if any
series_url	string	URL of the article series, if any
text	string	Prose text of the article (tables not included)
full_article	string	Complete article: prose followed by tables as markdown
inline_tables_md	string	Statistical tables from the article as markdown
noegletal_text	string	Key-figure callout text (e.g. Erhvervstillid 105,5 april 2026), if present
license	string	CC BY 4.0
source	string	Danmarks Statistik
tables	list	Raw StatBank API data for tables referenced in the article (see below)

tables field

Each entry in the tables list is a struct with:

Field	Type	Description
table_code	string	StatBank table identifier (e.g. ETILLID)
table_title	string	Human-readable table name
unit	string	Unit of measurement (e.g. Indeks, Pct.)
periods	list[string]	Time periods included (e.g. ["2026M04", "2026M03", "2026M02"])
table_rows	string	JSON-encoded list of data rows; each row is a dict with dimension columns, TID (time period), and INDHOLD (value)

Usage

from datasets import load_dataset

ds = load_dataset("oliverkinch/dst-tables", split="train")

# Full article (prose + tables)
article = ds[0]
print(article["full_article"])

# Just the tables as markdown (e.g. as model input)
print(article["inline_tables_md"])

# Raw StatBank data
import json
for table in article["tables"]:
    print(table["table_code"], "-", table["table_title"])
    rows = json.loads(table["table_rows"])
    print(rows[:2])

Data Collection

Articles were scraped from dst.dk. Inline statistical tables (<table class="TabelSmal">) were extracted from the article HTML and converted to markdown. Raw StatBank data was fetched from api.statbank.dk using table codes found in article hyperlinks, with time periods limited to those available at the article's publication date.