A small FastAPI service, built in August 2022, that scores a piece of creative writing for fluency, flexibility and originality. Topics are found with BERTopic over sentence-transformer embeddings; the metrics are computed from the topic assignments. It was built alongside the adaptive writing-support study below.
Understanding Revision Behavior in Adaptive Writing Support Systems for Education
Luca Mouchel, Thiemo Wambsganß, Paola Mejia-Domenzain, Tanja Käser. EDM 2023, 2023.
DOI · Preprint · Abstract and BibTeX
creativity/
app.py FastAPI app: `/models` loads the models, `/creativity/` scores a text
load_model.py downloads and caches the models
topics.py BERTopic pipeline
metrics.py fluency, flexibility, originality
models/ empty; models are fetched at first run
src/train.py trains a topic model
dockerfile Procfile requirements.txt environment.yml
pip install -r requirements.txt
uvicorn creativity.app:appThe interactive docs are at http://127.0.0.1:8000/docs. A dockerfile builds the same service; the old deployment notes for a specific cloud registry were removed from this README.
Note: an early commit added a 90 MB topic model to the repository. It is no longer tracked, but it is still in the history, so clones are large.
No data is included.
@inproceedings{mouchel2023understand,
title = {{Understanding Revision Behavior in Adaptive Writing Support Systems for Education}},
author = {Luca Mouchel and Thiemo Wambsganß and Paola Mejia-Domenzain and Tanja Käser},
year = {2023},
booktitle = {Zenodo},
publisher = {European Organization for Nuclear Research},
doi = {10.5281/zenodo.8115766},
eprint = {2306.10304},
archivePrefix = {arXiv},
url = {https://paola-md.github.io/papers/understanding-revision-behavior-in-adaptive-writing-support-systems-for-.html}
}No licence file. Contact me before reusing the code.
Not maintained: kept as the record of the paper.