NAMAA-Space/Arabic-Triplet-With-Multi-Negatives

---
license: apache-2.0
task_categories:
- feature-extraction
- sentence-similarity
language:
- ar
size_categories:
- 10K<n<100K
---

# Arabic Triplet with Multi Negatives

## Dataset Summary

This dataset is a modified version of the Arabic subset of the [Mr. TyDi dataset](https://huggingface.co/datasets/castorini/mr-tydi), tailored for retrieval and re-ranking tasks. The original dataset has been restructured by splitting the negative passages into separate fields (`negative1`, `negative2`, ..., `negativeN`) for each query. This modification allows more flexibility for training and evaluating retrieval and re-ranking models.

The dataset retains the original intent of Mr. Tydi, focusing on monolingual retrieval for the Arabic language while offering a new structure for ease of use in ranking-based learning tasks.

## Dataset Structure

The dataset includes train split only where each query is paired with a set of positive passages and multiple individually enumerated negative passages (up to 30).

### Example Data

#### Train Set

```json
{
"query_id": "1",
"query": "متى تم تطوير نظرية الحقل الكمي؟",
"positive_passages": [
{
"text": "بدأت نظرية الحقل الكمي بشكل طبيعي بدراسة التفاعلات الكهرومغناطيسية ..."
}
],
"negative1": {
"text": "تم تنفيذ النهج مؤخرًا ليشمل نسخة جبرية من الحقل الكمي ..."
},
"negative2": {
"text": "تتناول هذه المقالة الخلفية التاريخية لتطوير نظرية الحقل ..."
},
...
}
```

### Language Coverage
The dataset focuses exclusively on the **Arabic** subset of Mr. TyDi.

### Loading the Dataset

You can load the dataset using the **datasets** library from Hugging Face:

```python
from datasets import load_dataset

dataset = load_dataset('NAMAA-Space/Arabic-Triplet-With-Multi-Negatives')
dataset
```

### Dataset Usage

The new format facilitates training retrieval and re-ranking models by providing explicit negative passage fields. This structure simplifies the handling of negative examples during model training and evaluation.

### Citation Information

If you use this dataset in your research, please cite the original Mr. TyDi paper and this dataset as follows:

```
@article{mrtydi,
title={{Mr. TyDi}: A Multi-lingual Benchmark for Dense Retrieval},
author={Xinyu Zhang and Xueguang Ma and Peng Shi and Jimmy Lin},
year={2021},
journal={arXiv:2108.08787},
}

@dataset{Namaa,
title={Arabic Triplet With Multi Negatives},
author={Omer Nacar},
year={2024},
note={Hugging Face Dataset Repository}
}
```