---
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
dataset_info:
  features:
  - name: text
    dtype: string
  - name: labels
    sequence: string
  - name: id
    dtype: int64
  - name: title
    dtype: string
  splits:
  - name: train
    num_bytes: 415213673.5927568
    num_examples: 283808
  - name: test
    num_bytes: 103804881.40724319
    num_examples: 70953
  download_size: 284679882
  dataset_size: 519018555
license: mit
task_categories:
- text-classification
language:
- da
tags:
- chemistry
- biology
- finance
- legal
- music
- art
- webdataset
---
# Danish Wikipedia Categories

The dataset was created entirely from the [last Danish Wikipedia dump](https://dumps.wikimedia.org/dawiki/20240101/)
by traversing the category hierarchy in the [categorylinks table](https://www.mediawiki.org/wiki/Manual:Categorylinks_table).

All categories that were one level bellow the topcategories, and which had more than 30 articles assigned to them were selected.
In order to see whether an article belongs to a certain category I checked, whether the article was connected to the category in the directed graph of the category hierarchy.
If the length of the path between the article and a category was less than or equal to 5 edges (arbitrary choice, but I found it to work well),
the category was assigned to the article.

Note that this is a multilabel classification task.
Its purpose is to train sentence transformers for the [DFM project](https://centre-for-humanities-computing.github.io/danish-foundation-models/) using a technique similar to [SetFit](https://arxiv.org/abs/2209.11055).