project: legalis

• project around court case prediction from facts to binary outcome and tenor
• using heavily processed German court case data from OpenLegalData.io
• currently WIP

About the Project

Legalis is a University project for a machine learning and data science course. I wrote a paper at the University of Oslo about court case outcome prediction and this in the continuation of the project.

I'm using bulk data from openlegaldata, which included about 250.000 cases, out of which ca. 38k are usable for me.

Technology & Tools

I very much enjoy all the features 🤗 huggingface provides and heavily rely on it in my machine learning projects. Especially the hosting of Dataset, Model and Apps/Spaces. Additionally, I will use sckit-learn and the ChatGPT API.

ChatGPT works great for the extraction of certain information from longer text (if you're willing to pay or have short texts). I originally planned to use it for processing 30.000 cases, but that is quite pricey.

Working Steps

I had to do extensive preprocessing and cleaning of the data since the json dumps only provide the entire text in html format. After lots of work and cleaning I was able to extract 2800 cases to include a tenor (summary), reasoning and facts. Because of ChatGPT pricing I cannot extract facts from the text of the other 30.000 cases (it does work though).

This I will use to train to extract a binary ruling with ChatGPT and then train a sckit model to predict the outcome. We'll see about the accuracy of that.