---
license: mit
task_categories:
- text-generation
- information-extraction
language:
- en
tags:
- synthetic
- dates
- cultural-heritage
- museum
size_categories:
- n<1K
---

# synth-creation-date

## Dataset Description

This is a synthetic dataset for training date extraction models on cultural heritage and museum object descriptions. The dataset contains 792 samples of text descriptions paired with structured date information.

## Dataset Structure

### Data Fields

- `prompt`: Input text containing date information (string)
- `completion`: JSON string containing extracted dates with the following structure:
  ```json
  {
    "dates": [
      {
        "type": "creation_start|creation_end|destruction_start|destruction_end|other",
        "year": integer or null,
        "month": integer or null, 
        "day": integer or null,
        "reasoning": "brief explanation",
        "date_text": "exact phrase from text"
      }
    ]
  }
  ```

### Date Types

- `creation_start`: When production/creation began
- `creation_end`: When production/creation was completed  
- `destruction_start`: When destruction/damage began
- `destruction_end`: When destruction/damage was complete
- `other`: Other dates (acquisition, discovery, exhibition, etc.)

## Usage

This dataset is designed for training models to extract structured date information from cultural object descriptions.

```python
from datasets import load_dataset

dataset = load_dataset("yale-cultural-heritage/synth-creation-date")
```

## Generation Details

- Generated using Faker for structured date data
- Object types: 50+ different museum objects (paintings, sculptures, vases, etc.)
- Countries: Randomized using Faker
- Formats: HTML, JSON, and natural language descriptions
- Date ranges: 1300-2024 (with 10% generic centuries/decades)

## License

MIT License