Omartificial-Intelligence-Space
commited on
Update readme.md
Browse files
README.md
CHANGED
@@ -18,4 +18,84 @@ configs:
|
|
18 |
data_files:
|
19 |
- split: train
|
20 |
path: data/train-*
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
data_files:
|
19 |
- split: train
|
20 |
path: data/train-*
|
21 |
+
license: apache-2.0
|
22 |
+
task_categories:
|
23 |
+
- sentence-similarity
|
24 |
+
- feature-extraction
|
25 |
+
language:
|
26 |
+
- ar
|
27 |
+
size_categories:
|
28 |
+
- 100K<n<1M
|
29 |
---
|
30 |
+
|
31 |
+
# Arabic Mr. TyDi in Triplet Format
|
32 |
+
|
33 |
+
## Dataset Summary
|
34 |
+
|
35 |
+
This dataset is a transformed version of the Arabic subset of the [Mr. TyDi dataset](https://huggingface.co/datasets/castorini/mr-tydi), designed specifically for training retrieval and re-ranking models. Each query is paired with a positive passage and one of the multiple negative passages in a triplet format: `(query, positive, negative)`. This restructuring resulted in a total of 362,000 rows, making it ideal for pairwise ranking tasks and contrastive learning approaches.
|
36 |
+
|
37 |
+
The dataset maintains the original purpose of Mr. TyDi for monolingual retrieval, while offering a simplified and scalable format for learning-to-rank tasks.
|
38 |
+
|
39 |
+
## Dataset Structure
|
40 |
+
|
41 |
+
The dataset includes a train split, presented in the triplet format with the following fields:
|
42 |
+
|
43 |
+
- `query`: The query string.
|
44 |
+
- `positive`: The relevant passage for the query.
|
45 |
+
- `negative`: A non-relevant passage for the query.
|
46 |
+
|
47 |
+
### Example Data
|
48 |
+
|
49 |
+
#### Triplet Format
|
50 |
+
|
51 |
+
```json
|
52 |
+
{
|
53 |
+
"query": "متى تم تطوير نظرية الحقل الكمي؟",
|
54 |
+
"positive": {
|
55 |
+
"text": "بدأت نظرية الحقل الكمي بشكل طبيعي بدراسة التفاعلات الكهرومغناطيسية ..."
|
56 |
+
},
|
57 |
+
"negative": {
|
58 |
+
"text": "تم تنفيذ النهج مؤخرًا ليشمل نسخة جبرية من الحقل الكمي ..."
|
59 |
+
}
|
60 |
+
}
|
61 |
+
```
|
62 |
+
|
63 |
+
### Language Coverage
|
64 |
+
The dataset focuses exclusively on the **Arabic** subset of Mr. TyDi.
|
65 |
+
|
66 |
+
### Loading the Dataset
|
67 |
+
|
68 |
+
You can load the dataset using the **datasets** library from Hugging Face:
|
69 |
+
|
70 |
+
```python
|
71 |
+
from datasets import load_dataset
|
72 |
+
|
73 |
+
dataset = load_dataset('NAMAA-Space/Ara-TyDi-Triplet')
|
74 |
+
dataset
|
75 |
+
```
|
76 |
+
|
77 |
+
|
78 |
+
### Dataset Usage
|
79 |
+
|
80 |
+
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.
|
81 |
+
|
82 |
+
### Citation Information
|
83 |
+
|
84 |
+
If you use this dataset in your research, please cite the original Mr. TyDi paper and this dataset as follows:
|
85 |
+
|
86 |
+
```
|
87 |
+
@article{mrtydi,
|
88 |
+
title={{Mr. TyDi}: A Multi-lingual Benchmark for Dense Retrieval},
|
89 |
+
author={Xinyu Zhang and Xueguang Ma and Peng Shi and Jimmy Lin},
|
90 |
+
year={2021},
|
91 |
+
journal={arXiv:2108.08787},
|
92 |
+
}
|
93 |
+
|
94 |
+
@dataset{Namaa,
|
95 |
+
title={Ara TyDi Triplet},
|
96 |
+
author={Omer Nacar},
|
97 |
+
year={2024},
|
98 |
+
note={Hugging Face Dataset Repository}
|
99 |
+
}
|
100 |
+
```
|
101 |
+
|