metadata
dataset_info:
features:
- name: s1
dtype: string
- name: s2
dtype: string
splits:
- name: train
num_bytes: 26894269
num_examples: 131157
download_size: 12694309
dataset_size: 26894269
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
task_categories:
- sentence-similarity
language:
- fa
pretty_name: PersianSimilarSentences
size_categories:
- 100K<n<1M
Dataset Summary
PersianSimilarSentences is a curated dataset of Persian sentence pairs designed for training semantic similarity and sentence embedding models. It combines several existing Persian NLP resources and a machine-translated version of the Quora Question Pairs dataset.
This dataset was created to fine-tune the FaLaBSE and FaMiniLM models as part of the paper "MetaRAG and WikiFaQA: A Co-designed Framework and Benchmark for Advancing Persian Long-Context RAG."
Citation
@unpublished{mobarekati2024metarag,
title={MetaRAG and WikiFaQA: A Co-designed Framework and Benchmark for Advancing Persian Long-Context RAG},
author={Mobarekati, Ali and Mohades, Ali},
note={Unpublished manuscript},
year={2025}
}