|
--- |
|
task_categories: |
|
- text-classification |
|
language: |
|
- ba |
|
- kk |
|
- ky |
|
size_categories: |
|
- 10K<n<100K |
|
--- |
|
|
|
# Triplets for Turkic languages language models |
|
|
|
## Description |
|
This dataset is designed to test models for working with Next Sentence Prediction (NSP) and Sentence Order Prediction (SOP). It includes two sub-sets with triplets of texts.. |
|
|
|
## Usage |
|
This dataset can be used to train and evaluate models capable of performing NSP and SAP tasks. |
|
|
|
## Dataset structure: |
|
Each entry in the dataset represents three values: |
|
- **text**: a triplet of text; |
|
- **flag**: a flag indicating whether the order of sentences in this triplet is correct; |
|
- **lang**: the language of the sentence. |
|
|
|
## The creation process |
|
Using the functions described on [github](https://github.com/Electrotubbie/turk_langs_analyse) preprocessing and analysis of texts from [dataset](https://huggingface.co/datasets/Electrotubbie/classification_Turkic_languages) was performed and triplets are selected according to certain rules (triplets should be approximately the same length and from 30 to 100 characters). |
|
Also, the triplets were selected in such a way that no sentence displayed in the dataset was repeated several times. |
