v_1.0

.gitattributes

@@ -58,7 +58,3 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
-v1.0.jsonl filter=lfs diff=lfs merge=lfs -text
-v1.0.parquet filter=lfs diff=lfs merge=lfs -text
-v1.1.jsonl filter=lfs diff=lfs merge=lfs -text
-v1.1.parquet filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,84 +1,3 @@
 ---
 license: mit
-task_categories:
-- text-classification
-language:
-- fr
-- de
-- it
-tags:
-- entailment
-- politics
-- switzerland
-- voting
-- multilingual
-pretty_name: MNLI over Swiss Voting Booklets
-size_categories:
-- 1K<n<10K
 ---
-# MNLI over Swiss Voting Booklets: Benchmark Dataset  
-
-## Dataset Contents  
-
-### Fields  
-
-|        **Field Name**   | **Data Type** | **Possible Content** | **Description**                                                                                                                                                                                                                                                                                  |
-|-------------------------|---------------|----------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
-| `claim`                 | `string`      | Raw Text             | Claim about the contents of a voting booklet. May be entailed, contradictory or unrelated to its source material.                                                                                                                                                                                |
-| `claim_language`        | `string`      | [`de`, `fr`, `it`]   | The language the claim is written in. May be German (`de`), French (`fr`) or Italian (`it`). Does not need to match the `reference_language`.                                                                                                                                                    |
-| `reference_string`      | `string`      | Raw Text             | A text snippet from the voting booklet that was used to generate the `claim`.                                                                                                                                                                                                                    |
-| `entailment_label`      | `int`         | [`0`, `1`, `2`]      | The entailment class of the `claim` relative to the `reference_string` and voting booklet as a whole. May be `Entailment` (`0`), `Unrelated / Neutral` (`1`) or `Contradiction` (`2`).                                                                                                           |
-| `baseline_score`        | `float`       | [`0.0`,...,`1.0`]    | The certainty about the label correctness of the classifier used for validation.                                                                                                                                                                                                                 |
-| `reference_language`    | `string`      | [`de`, `fr`, `it`]   | The language the `reference_string` and corresponding voting booklet is written in. May be German (`de`), French (`fr`) or Italian (`it`). Does not need to match the `claim_language`.                                                                                                          |
-| `booklet_publish_date`  | `string`      | `YYYY-MM-DD`         | The date at which the voting booklet was published.                                                                                                                                                                                                                                              |
-| `booklet_download_date` | `string`      | `YYYY-MM-DD`         | The date at which the booklet was downloaded for the creation of the dataset.                                                                                                                                                                                                                    |
-| `booklet_url`           | `string`      | `URL`                | Download link for the full booklet as a PDF.                                                                                                                                                                                                                                                     |
-| `vote`                  | `string`      | Raw Text             | Name of the vote the `claim` is about. Can be used to verify ambiguous claims about a booklet containing multiple votes, e.g. `Der Bundesrat empfiehlt, die Initiative abzulehnen.` (`The federal council recommends to reject the initiative.`) when the booklet contains multiple initiatives. |
-
-### Sampling Ratios  
-
-To create the dataset, we sampled a larger corpus of `claim` / `reference_string` pairs for rows fulfilling specific criteria to ensure that the resulting dataset is balanced.  
-We specifically designed the dataset to target...
-
-- ... a uniform distribution over the `claim_language` field.  
-- ... a uniform distribution over the `entailment_label` field.  
-- ... an approximately `0.5 / 0.35 / 0.15` split of `easy / medium / hard` claims to classify, based on the `baseline_score` provided by our validation model.  
-
-## How to Use the Dataset (Quick Start in Python)  
-
-Install the `datasets` package with a package manager of your choice. E.g. `uv` on Linux:
-
-```console
-> mkdir my-project
-> cd my-project/
-> uv init
-> uv add datasets
-```
-
-You can then download and use this dataset with the `load_dataset` function as such:
-
-```python
-from datasets import load_dataset
-
-dataset = load_dataset("OSTswiss/MNLIoverSwissVotingBooklets")
-print(dataset)
-```
-
-For more information or alternative approaches, consult the [Downloading datasets](https://huggingface.co/docs/hub/en/datasets-downloading) page by Hugging Face.
-
-## Additional Information  
-
-### Hack Apertus  
-
-This dataset was created as part of the `Hack Apertus` hackathon series. If you want to participate or contribute, you can find more information on [https://hackapertus.ch/](https://hackapertus.ch/).  
-
-### Credit  
-
-This Dataset was created by the `AI Applications and Deployment Lab`, part of the `Institute for Software (IFS)` at `Eastern Switzerland University of Applied Sciences OST` ([Website](https://www.ost.ch/en/research-and-consulting-services/computer-science/ifs-institute-for-software/ai-applications-and-deployment-lab)).  
-
-Special thanks go out to the Development Team:  
-
-- Prof. Dr. Mitra Purandare, `mitra.purandare@ost.ch`, [ Project Lead, Hackathon Task Design ]  
-- Abinas Kuganathan, `abinas.kuganathan@ost.ch`, [ Hackathon Task Design, Programming ]  
-- Dominik Castelberg, `dominik.castelberg@ost.ch`, [ Dataset Design, Programming ]  
-- Vladimir Zivanovic, `vladimir.zivanovic@ost.ch`, [ Programming ]

v1.1.jsonl

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:87ee4afd9c0ee8861106c50f05884481402ca668aecf927df68fc52d915b02fd
-size 12742710

v1.1.parquet

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:444fe113115a1fa4134664f67ce572cb04b170b443bc741ae546a34c2f76b7c7
-size 2736723