The underlying corpus was trained on the basis of over 300,000 texts of the mentioned language categories. Freely available websites served as sources. Thematic diversity was taken into account when selecting the sources.

Developed by: Ruben Klepp
Language(s) (NLP): de
License: apache-2.0
Parent Model: https://huggingface.co/deepset/gbert-base

Evaluation

f1: 0.982
Precision: 0.981
Recall: 0.983

How to use

python

>>> from transformers import pipeline
>>> classifier = pipeline(model="krupper/text-complexity-classification")
>>> classifier("Bei Kleinkindern unter 2 Jahren liegen nur begrenzte Erfahrungen zur Pharmakokinetik vor.")
[{'label': 'special_language', 'score': 0.999923825263977}]

DOI: https://doi.org/10.57967/hf/0131