---
license: mit
dataset_info:
  features:
  - name: id
    dtype: int64
  - name: court
    struct:
    - name: id
      dtype: int64
    - name: jurisdiction
      dtype: string
    - name: level_of_appeal
      dtype: string
    - name: name
      dtype: string
    - name: state
      dtype: int64
  - name: file_number
    dtype: string
  - name: date
    dtype: timestamp[s]
  - name: type
    dtype: string
  - name: content
    dtype: string
  - name: tenor
    dtype: string
  - name: facts
    dtype: string
  - name: reasoning
    dtype: string
  splits:
  - name: three
    num_bytes: 169494251
    num_examples: 2828
  - name: two
    num_bytes: 183816899
    num_examples: 4954
  download_size: 172182482
  dataset_size: 353311150
task_categories:
- text-classification
language:
- de
tags:
- legal
pretty_name: Edited German Court case decision
size_categories:
- 1K<n<10K
---

# Dataset Card for openlegaldata.io bulk case data

## Dataset Description

This is a edit/cleanup of Bulk Data of [openlegaldata.io](https://de.openlegaldata.io/), which I also brought onto Huggingface [here](LennardZuendorf/openlegaldata-bulk-data).

#### The Entire Dataset Is In German

- **Github Repository:** [uniArchive-legalis]](https://github.com/LennardZuendorf/uniArchive-legalis)
- **Repository:** [Bulk Data](https://static.openlegaldata.io/dumps/de/)

## Edit Summary

I have done some cleaning and splitting of the data and filtered out large parts that were not (easily) usable, cutting down the number of cases to at max 4000 - from 250000. This results in two different splits. Which is because German Courts don't format their case decision the same way.

### Data Fields

Independent of the split, most fields are the same, they are:

| id | court | file_number | date | type | content
| - | - | - | - | - | - |
| numeric id | name of the court that made the decision | file number of the case ("Aktenzeichen") | decision date | type of the case decision | entire content (text) of the case decision

Additionally, I added 3 more fields because of the splitting of the content:

#### Two Split

- Case Decision I could split into two parts: tenor and reasoning.
- Which means the three fields tenor, content and facts contain the following:

| tenor | reasoning | facts
| - | - | - |
| An abstract, legal summary of the cases decision | the entire rest of the decision, explaining in detail why the decision has been made | an empty text field |

#### Three Split

- Case Decision I could split into three parts: tenor, reasoning and facts
- This Data I have used to create binary labels with the help of ChatGPT, see [legalis](https://huggingface.co/datasets/LennardZuendorf/legalis) for that
- The three fields tenor, content and facts contain the following:

| tenor | reasoning | facts
| - | - | - |
| An abstract, legal summary of the cases decision | the entire rest of the decision, explaining in detail why the decision has been made | the facts and details of a case |

### Languages

- German

## Additional Information

### Licensing/Citation Information

The [openlegaldata platform](https://github.com/openlegaldata/oldp) is licensed under the MIT license, you can access the dataset by citing the original source, [openlegaldata.io](https://de.openlegaldata.io/) and me, [Lennard Zündorf](https://github.com/LennardZuendorf) as the editor of this dataset.