Code for "GerPS-Compare: Comparing NER methods for legal norm analysis"

This repository contains all code for the comparison of different models for supporting legal norm analysis with different categories. 

The directories for the approaches contain the code to replicate the training and prediction.
The Evaluation repository contains the code to perform the evaluation of the files that are created during the prediction / application.

So to replicate our results, do the following:
1. read the README.md for each approach and follow its instructions for training and prediction / application
2. read the README.md in the `.\Evaluation` directory and follow the instructions

The `.\Corpus` directory contains information about the modifications that we applied to the GerPS-NER corpus [1] and also contains the final splits that we used for training, development and evaluation.
It contains the final splits in two forms: one where all sentences are in one file, and one where each split is a folder with one file per sentence. This was necessary due to the different input structures of the models.

[1] Feddoul, L., Bachinger, S. T., Apel, S., Karg, P., Klewer, N., Forshayt, D., Erd, R., Mauch, M., & Lachenmaier, C. (2024). GerPS-NER: Dataset and code [Data set]. Zenodo. https://doi.org/10.5281/zenodo.10822682Feddoul, L., Bachinger, S. T., Apel, S., Karg, P., Klewer, N., Forshayt, D., Erd, R., Mauch, M., & Lachenmaier, C. (2024). GerPS-NER: Dataset and code [Data set]. Zenodo. https://doi.org/10.5281/zenodo.10822682