---
paper: arxiv:2109.03794
size_categories:
- n<1K
task_categories:
- token-classification
language:
- en
tags:
- pipeline-numbers
- ner
- p-and-id
---

# Digitize-PID: Pipeline numbers (NER)

**Note**: *I am not the author of this dataset*

Named Entity Recognition dataset for extracting pipeline numbers from full text of P&ID 
(Piping and Instrumentation Diagram) documents.

## Dataset Details

### Dataset Description

Pipeline numbers are structured identifiers in engineering documents:
- Example Format: `A-123-BC` (3-5 segments with a separator such as `-`, ` `, or `_`)
- Use case: Automated extraction from P&ID document text
- Domain: Process and piping industry

### Data Fields

- `id`: Unique example identifier
- `tokens`: List of tokenized words/punctuation
- `labels`: BIO tags for each token
- `pipeline_numbers`: Ground truth pipeline numbers
- `full_text`: Original text

### Label Schema

| Label | Meaning |
|-------|---------|
| `B-PIPE` | Beginning of pipeline number |
| `I-PIPE` | Inside pipeline number |
| `O` | Outside (not pipeline number) |

### Splits

Data was randomly split.

| Split | Examples |
|-------|----------|
| train |      400 |
| validation | 50  |
| test |       50  |

### Data Creation

- **Source:** Digitize-PID
- **Annotation:** Automatic BIO tagging with character-level alignment

## Usage

### With Hugging Face Datasets

```python
from datasets import load_dataset

dataset = load_dataset("hamzas/digitize-pid-ner")

print(dataset)

# Access example
example = dataset['train'][0]
print(example['tokens'])
print(example['labels'])
```