Topic modeling of Swiss Court Rulings

Research Center for Digital Sustainability (RCDS)

--

Renato Rao, Marius Asadauskas, Dominik Ummel

When performing various analysis on Court Rulings the vast amount of available data makes the analysis rather complex and time consuming. Therefore the goal of the project was to perform topic modeling over a multilingual dataset of Swiss Court Rulings to group the Rulings into meaningful topics, so that analysis can be done on these groupings.

This repository contains the colab notebook to traing the topic models. Furthermore, it contains the resulting visualizations and evaluations.

Dataset generation

The dataset sizes are approximately:

• Germam: 5GB
• French: 3GB
• Italian: .5GB

To generate the dataset on the sandbox (kindly provided by RCDS) run:

psql -d scrc -U readonly  -c "\copy de(id,court,language,title,header,text,html_url,date) TO 'train_de.csv' DELIMITER ';' CSV HEADER;"

psql -d scrc -U readonly  -c "\copy fr(id,court,language,title,header,text,html_url,date) TO 'train_fr.csv' DELIMITER ';' CSV HEADER;"

psql -d scrc -U readonly  -c "\copy it(id,court,language,title,header,text,html_url,date) TO 'train_it.csv' DELIMITER ';' CSV HEADER;"

Important: Characters such as ü are stored as u + ¨ and not ü, so the dataset needs to be normalized

Colab Notebook

For an interactive session in the browser the jupyter notebook can be seen here:

https://drive.google.com/file/d/1pLeBaMzvvfXjjnsjTtw_sPvdpUWzIQkl/view?usp=sharing

Model download

The trained BERTopic model (~3GB) can be downloaded here:

https://drive.google.com/file/d/1fSAjlctonmHIN37iL0vFcesZc5MMQTt_/view?usp=sharing

Sources

[1]M. Grootendorst, ‘BERTopic: Leveraging BERT and c-TF-IDF to create easily interpretable topics.’ Zenodo, 2020. doi: 10.5281/zenodo.4381785. [2]F. Bianchi, S. Terragni, D. Hovy, D. Nozza, and E. Fersini, ‘Cross-lingual Contextualized Topic Models with Zero-shot Learning’, in Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, Online, Apr. 2021, pp. 1676–1683. [Online]. Available: https://www.aclweb.org/anthology/2021.eacl-main.143

Github Links:

https://github.com/MaartenGr/BERTopic

https://github.com/MilaNLProc/contextualized-topic-models