README.md

---
annotations_creators:
- other
language:
- en
language_creators:
- found
license:
- other
multilinguality:
- monolingual
size_categories:
- 10K<n<100k
source_datasets:
- extended|other
tags:
- relation extraction
- information-extraction
dataset_info:
  features:
  - name: text
    dtype: string
  - name: entity1
    dtype: string
  - name: entity2
    dtype: string
  - name: relation
    dtype: string
  - name: prompt_0_shot
    dtype: string
  - name: prompt_2_shot
    dtype: string
  - name: prompt_5_shot
    dtype: string
  splits:
  - name: train
    num_bytes: 67454455
    num_examples: 11297
  - name: validation
    num_bytes: 11182197
    num_examples: 1864
  - name: test
    num_bytes: 33938174
    num_examples: 5663
  download_size: 57694394
  dataset_size: 112574826
task_categories:
- text-classification
- text-generation
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---

# GIDS (Relation Extraction)

GIDS (**Google-IISc Distant Supervision**) is a **general-domain sentence-level relation
extraction (RE)** dataset. Each example pairs a sentence (mentioning a head and tail entity)
with the relation that holds between the two entities. It was built by extending the
human-judged **Google Relation Extraction** corpus with additional distantly-supervised
sentences, and is designed so that every entity pair has at least one sentence that expresses
the target relation (reducing the noise typical of distant supervision).

This copy is packaged for the paper **"Sub-Billion, Super-Frontier: Fine-Tuned Small Language
Models Rival Zero-Shot Frontier LLMs on General and Literary Relation Extraction"**
(Christou & Tsoumakas, 2026) [arXiv:2606.22606](https://arxiv.org/abs/2606.22606). Rows are
pre-formatted as instruction prompts so the dataset can be used directly for prompt-conditioned
fine-tuning and evaluation.

## Dataset structure

Splits: `train`, `validation`, `test`.

| Column | Description |
|---|---|
| `prompt_0_shot` | Zero-shot instruction prompt (task instructions + the input sentence). |
| `prompt_2_shot` | Same prompt with 2 in-context demonstrations prepended. |
| `prompt_5_shot` | Same prompt with 5 in-context demonstrations prepended. |
| `relation` | Gold relation label (the target/completion). |

The `relation` column holds a small set of Freebase-style relation types (e.g. birth/death place
and education-related relations) plus a no-relation / `NA` label; inspect the column for the exact
set. The three `prompt_*` columns are alternative renderings of the **same** example at different
shot counts, so pick one shot setting per experiment rather than concatenating them.


An example of 'train' looks as follows:
```json
{
  "text": "War as appropriate. Private Alfred James_Smurthwaite Sample. 26614. 2nd Battalion Yorkshire Regiment. Son of Edward James Sample, of North_Ormesby , Yorks. Died 2 April 1917. Aged 29. Born Ormesby, Enlisted Middlesbrough. Buried BUCQUOY ROAD CEMETERY, FICHEUX. Not listed on the Middlesbrough War Memorial Private Frederick Scott. 46449. 4th Battalion Yorkshire Regiment. Son of William and Maria Scott, of 25, Aspinall St., Heywood, Lancs. Born at West Hartlepool. Died 27 May 1918. Aged 24.", 
  "entity1": "Alfred", 
  "entity2": "Yorkshire", 
  "entity1Type": "PERSON", 
  "entity2Type": "LOCATION", 
  "relation": 'fight_at'
}
```


## Data Splits

|      | Train | Dev  | Test |
|------|-------|------|------|
| GIDS | 11297 | 1864 | 5663 |

## Usage

```python
from datasets import load_dataset

ds = load_dataset("Despina/gids")
print(ds["test"][0]["prompt_2_shot"])  # formatted input
print(ds["test"][0]["relation"])       # gold label
```

## Source and licensing

GIDS is derived from the **Google Relation Extraction** corpus (distantly extended over
Wikipedia text) and was introduced by Jat, Khandelwal & Talukdar (2018). It is available for
research use; please respect the terms of the underlying Google Relation Extraction corpus and
cite the original GIDS resource alongside the paper below.

## Citation

**If you use this dataset, please cite our paper:**

```bibtex
@article{christou2026subbillion,
  title        = {Sub-Billion, Super-Frontier: Small Language Models Rival
                  Zero-Shot Frontier LLMs on General and Literary Relation Extraction},
  author       = {Christou, Despina and Tsoumakas, Grigorios},
  journal      = {arXiv preprint arXiv:2606.22606},
  year         = {2026},
  url          = {https://arxiv.org/abs/2606.22606}
}
```