---
license: apache-2.0
task_categories:
- information-retrieval
- text-retrieval
tags:
- beir
- scidocs
- information-retrieval
- retrieval
- search
dataset_info:
- config_name: corpus
  features:
  - name: title
    dtype: string
  - name: text
    dtype: string
  - name: _id
    dtype: string
  - name: metadata
    struct:
    - name: authors
      list: string
    - name: cited_by
      list: string
    - name: references
      list: string
    - name: year
      dtype: int64
  splits:
  - name: train
    num_bytes: 255256970
    num_examples: 25657
  download_size: 219636587
  dataset_size: 255256970
- config_name: queries
  features:
  - name: text
    dtype: string
  - name: _id
    dtype: string
  - name: metadata
    struct:
    - name: authors
      list: string
    - name: cited_by
      list: string
    - name: references
      list: string
    - name: year
      dtype: int64
  splits:
  - name: train
    num_bytes: 3099119
    num_examples: 1000
  download_size: 2797538
  dataset_size: 3099119
configs:
- config_name: corpus
  data_files:
  - split: train
    path: corpus/train-*
- config_name: queries
  data_files:
  - split: train
    path: queries/train-*
---

# BEIR SCIDOCS Dataset (Migrated)

This is a migrated version of BeIR/scidocs that is compatible with datasets library 4.0.0+.

## Dataset Description

This dataset contains the scidocs dataset from the BEIR benchmark, converted from the old script-based format to Parquet format.

## Dataset Structure

### Queries

- **Split 'queries'**: 1,000 examples
  - Features: ['_id', 'text', 'metadata']

- **Total examples**: 1,000


### Corpus

- **Split 'corpus'**: 25,657 examples
  - Features: ['_id', 'title', 'text', 'metadata']

- **Total examples**: 25,657


## Usage

```python
from datasets import load_dataset

# Load queries (split: queries)
queries = load_dataset("Hyukkyu/beir-scidocs", "queries", split="queries")

# Load corpus (split: corpus)
corpus = load_dataset("Hyukkyu/beir-scidocs", "corpus", split="corpus")
```


## Available Splits

### Queries
- `queries`: 1,000 examples

### Corpus
- `corpus`: 25,657 examples

## Original Dataset

This dataset is migrated from: BeIR/scidocs

## Citation

If you use this dataset, please cite the original BEIR paper:

```bibtex
@article{thakur2021beir,
  title={BEIR: A Heterogeneous Benchmark for Zero-shot Evaluation of Information Retrieval Models},
  author={Thakur, Nandan and Reimers, Nils and Ruckle, Andreas and Srivastava, Abhishek and Gurevych, Iryna},
  journal={arXiv preprint arXiv:2104.08663},
  year={2021}
}
```