This repository contains the code for scraping every article from the Tagesschau.de archive. Find a deduplicated version of the dataset from 01.01.2018 to 26.04.2023 on HuggingFace.
CSV structure:
Field | Description |
---|---|
date |
Article date |
headline |
Article headline |
short_headline |
A short headline / Context |
description |
Article lead |
article_body |
Raw article text |
detailed_article_body |
Article body with text objects (subhead, time, paragraph, topline) |
date_modified |
Modified timestamp |
date_published |
Published timestamp |
keywords |
List of keywords |
author |
Author object |
publisher |
Publisher object |
label |
Article label (e.g. interview, audio) |
link |
Article link |
Size:
The final dataset (2018-today) contains 225202 articles from 1942 days. Of these articles only 21848 are unique (Tagesschau often keeps articles in circulation for ~1 month). The total download size is ~65MB.
Cleaning:
- Duplicate articles are removed
- Articles with empty text are removed
- Articles with empty short_texts are removed
- Articles, headlines and short_headlines are stripped of leading and trailing whitespace
More details in clean.py
.
Install with:
pip install poetry
poetry install --no-root
Run with:
python scrape_tagesschau.py #--start_date=2018-01-01 --end_date=2023-01-01 --output=tagesschau.csv