mFollowIR-parquet

# mFollowIR-parquet

This is a parquet version of the mFollowIR dataset that can be loaded directly with `load_dataset()`. The original dataset can be found at [jhu-clsp/mFollowIR](https://huggingface.co/datasets/jhu-clsp/mFollowIR).

## Dataset Structure

The dataset contains the following configurations for each language (fas, rus, zho):

### Configurations
- `qrels_og_[lang]`: Original relevance judgments (test split)
- `qrels_changed_[lang]`: Modified relevance judgments (test split)
- `corpus_[lang]`: Document collection
- `queries_[lang]`: Query set with instructions
- `top_ranked_[lang]`: Top ranked documents

## Usage

```python
from datasets import load_dataset

# Load a specific configuration
dataset = load_dataset("jhu-clsp/mFollowIR-parquet", "queries_fas") # or any other config

# Load multiple configurations
dataset = load_dataset("jhu-clsp/mFollowIR-parquet",
["queries_fas", "corpus_fas"])
```

## Citation

```bibtex
@article{weller2024mfollowir,
title="mFollowIR: a Multilingual Benchmark for Instruction Following in Information Retrieval",
author="Weller, Orion and Chang, Benjamin and Yang, Eugene and Yarmohammadi, Mahsa and Barham, Sam and MacAvaney, Sean and Cohan, Arman and Soldaini, Luca and Van Durme, Benjamin and Lawrie, Dawn",
journal="arXiv preprint TODO",
year="2024"
}
```

# mFollowIR-parquet
本数据集为mFollowIR数据集的Parquet格式版本，可直接通过`load_dataset()`函数加载。原始数据集可在[jhu-clsp/mFollowIR](https://huggingface.co/datasets/jhu-clsp/mFollowIR)获取。

## 数据集结构
本数据集针对波斯语（fas）、俄语（rus）、中文（zho）三种语言提供如下配置项：

### 配置项
- `qrels_og_[lang]`：原始相关性标注（测试集划分）
- `qrels_changed_[lang]`：修改后的相关性标注（测试集划分）
- `corpus_[lang]`：文档集合
- `queries_[lang]`：带指令的查询集合
- `top_ranked_[lang]`：Top排序文档

## 用法
python
from datasets import load_dataset

# 加载指定配置项
dataset = load_dataset("jhu-clsp/mFollowIR-parquet", "queries_fas") # 或其他任意配置项

# 加载多个配置项
dataset = load_dataset("jhu-clsp/mFollowIR-parquet",
["queries_fas", "corpus_fas"])

## 引用
bibtex
@article{weller2024mfollowir,
title="mFollowIR: a Multilingual Benchmark for Instruction Following in Information Retrieval",
author="Weller, Orion and Chang, Benjamin and Yang, Eugene and Yarmohammadi, Mahsa and Barham, Sam and MacAvaney, Sean and Cohan, Arman and Soldaini, Luca and Van Durme, Benjamin and Lawrie, Dawn",
journal="arXiv preprint TODO",
year="2024"
}