Named Entity Recognition in Historical Czech Texts

This repository contains materials for a project focused on Named Entity Recognition (NER) in historical Czech documents, particularly digitized newspapers and chronicles.

Data

The experiments are based on several datasets:

• CNEC (Czech Named Entity Corpus) – reference dataset for Czech NER.
• Historical newspaper corpus – OCR-processed historical Czech texts.
• SYN v4 – large corpus of written Czech used for supplementary language modeling.

Dataset generator

First unzip both datasets inside resources/ After that run following command: python3 -m src.data.dataset_generator

Fine tune RobeCzech + test

To fine tune the model, run following command: python3 -m src.models.finetuned_robeczech_cnec

After that you can run python3 -m src.models.finetuned_robeczech_cnec_test to see evaluation results.

Inter Anotator Agreement

Run python3 get_iaa.py inside experiments folder.

Authors

Bc. Dominik Hofman

Bc. Lucie Jadrná

Bc. Vítězslav Cupl

Supervisor: Ing. Marek Vaško