Can LLMs Create Legally Relevant Summaries and Analyses of Videos?

This notebook contains the code for the paper:

Lyra Hoeben-Kuil, Gijs van Dijck, Jaromir Savelka, Johanna Gunawan, Konrad Kollnig, Marta Kołacz, Mindy Duffourc, Shashank M. Chakravarthy, Hannes Westermann. 2025. Can LLMs Create Legally Relevant Summaries and Analyses of Videos? In proceedings of JURIX 2025 (forthcoming)

The notebook contains the code necessary to iterate through a .CSV file of videos. In the first part of the notebook, the model will create a summary of the video, corresponding to RQ1 and RQ2 of the paper. In the second section, the model aims to connect these facts to the legally relevant rules, corresponding to RQ3.

Running the experiments

On Google Colab

The notebook is run most easily in Google Colab, available here. In colab, a secret should be set with the key "GOOGLE_API_KEY", containing the google API key. The code will clone the repository to load the data.

Locally

To run the experiments locally, first create a new .env file and add your google API key. Then, you can run the project using uv. After installing uv, simply start a jupyter notebook using:

uv run --with jupyter jupyter lab

Citation

If you use this code, please cite us at:

@inproceedings{HoebenKuil2025LLMs,
  author       = {Lyra Hoeben-Kuil and Gijs van Dijck and Jaromir Savelka and Johanna Gunawan and Konrad Kollnig and Marta Kołacz and Mindy Duffourc and Shashank M. Chakravarthy and Hannes Westermann},
  title        = {Can LLMs Create Legally Relevant Summaries and Analyses of Videos?},
  booktitle    = {Proceedings of the 38th International Conference on Legal Knowledge and Information Systems (JURIX 2025)},
  year         = {2025},
  note         = {Forthcoming},
}