---
configs:
- config_name: ner
  data_files: ner.csv
- config_name: el
  data_files: el.csv
- config_name: re
  data_files: re.csv
---

# Dataset Card for Text2Tech Curated Documents

## Dataset Summary

This dataset is the result of converting a UIMA CAS 0.4 JSON export from the Inception annotation tool into a simplified format suitable for Natural Language Processing tasks. Specifically, it provides configurations for Named Entity Recognition (NER), Entity Linking (EL), and Relation Extraction (RE).

The conversion process utilized the `dkpro-cassis` library to load the original annotations and `spaCy` for tokenization. The final dataset is structured similarly to the DFKI-SLT/mobie dataset to ensure compatibility and ease of use with the Hugging Face ecosystem.

This version of the dataset loader provides configurations for:

* **Named Entity Recognition (ner)**: NER tags use spaCy's BILUO tagging scheme.
* **Entity Linking (el)**: Entity mentions are linked to external knowledge bases.
* **Relation Extraction (re)**: Relations between entities are annotated.

## Supported Tasks and Leaderboards

* **Tasks**: Named Entity Recognition, Entity Linking, Relation Extraction
* **Leaderboards**: More Information Needed

## Languages

The text in the dataset is in English.

## Dataset Structure

### Data Instances

#### ner

An example of 'train' looks as follows.

```json
{
  "docid": "138",
  "tokens": [
    "\"",
    "Samsung",
    "takes",
    "aim",
    "at",
    "blood",
    "pressure",
    "monitoring",
    "with",
    "the",
    "Galaxy",
    "Watch",
    "Active",
    "..."
  ],
  "ner_tags": [
    0,
    1,
    0,
    0,
    0,
    2,
    3,
    4,
    0,
    0,
    5,
    6,
    7,
    "..."
  ]
}
```

#### el

An example of 'train' looks as follows.

```json
{
  "docid": "138",
  "tokens": [
      "\"",
      "Samsung",
      "takes",
      "aim",
      "at",
      "blood",
      "pressure",
      "monitoring",
      "with",
      "the",
      "Galaxy",
      "Watch",
      "Active",
       "..."
  ],
    "ner_tags": [
      0,
      1,
      0,
      0,
      0,
      2,
      3,
      4,
      0,
      0,
      5,
      6,
      7,
      "..."
],

  "entity_mentions": [
    {
      "text": "Samsung",
      "start": 1,
      "end": 2,
      "char_start": 1,
      "char_end": 8,
      "type": 0,
      "entity_id": "http://www.wikidata.org/entity/Q124989916"
    },
    "..."
  ]
}
```

#### re

An example of 'train' looks as follows.

```json
{
  "docid": "138",
  "tokens": [
    "\"",
    "Samsung",
    "takes",
    "aim",
    "at",
    "blood",
    "pressure",
    "monitoring",
    "with",
    "the",
    "Galaxy",
    "Watch",
    "Active",
    "..."
  ],
  "ner_tags": [
    0,
    1,
    0,
    0,
    0,
    2,
    3,
    4,
    0,
    0,
    5,
    6,
    7,
    "..."
  ],
  "relations": [
    {
      "id": "138-0",
      "head_start": 706,
      "head_end": 708,
      "head_type": 2,
      "tail_start": 706,
      "tail_end": 708,
      "tail_type": 2,
      "type": 0
    },
    "..."
  ]
}
```

### Data Fields

#### ner

* `docid`: A `string` feature representing the document identifier.
* `tokens`: A `list` of `string` features representing the tokens in the document.
* `ner_tags`: A `list` of classification labels using spaCy's BILUO tagging scheme. The mapping from ID to tag is as follows:

**BILUO Tagging Scheme:**
- **B-** (Begin): First token of a multi-token entity
- **I-** (Inside): Inner tokens of a multi-token entity  
- **L-** (Last): Final token of a multi-token entity
- **U-** (Unit): Single token entity
- **O** (Outside): Non-entity token

```json
{
  "O": 0,
  "U-Organization": 1,
  "B-Method": 2,
  "I-Method": 3,
  "L-Method": 4,
  "B-Technological System": 5,
  "I-Technological System": 6,
  "L-Technological System": 7,
  "U-Technological System": 8,
  "U-Method": 9,
  "B-Material": 10,
  "L-Material": 11,
  "I-Material": 12,
  "B-Organization": 13,
  "L-Organization": 14,
  "I-Organization": 15,
  "U-Material": 16,
  "B-Technical Field": 17,
  "L-Technical Field": 18,
  "I-Technical Field": 19,
  "U-Technical Field": 20
}
```

#### el

* `docid`: A `string` feature representing the document identifier.
* `tokens`: A `list` of `string` features representing the tokens in the document.
* `entity_mentions`: A `list` of `struct` features containing:
   * `text`: a `string` feature.
   * `start`: token offset start, a `int32` feature.
   * `end`: token offset end, a `int32` feature.
   * `char_start`: character offset start, a `int32` feature.
   * `char_end`: character offset end, a `int32` feature.
   * `type`: a classification label. The mapping from ID to entity type is as follows:

```json
{
  "Organization": 0,
  "Method": 1,
  "Technological System": 2,
  "Material": 3,
  "Technical Field": 4
}
```

   * `entity_id`: a `string` feature representing the entity identifier from a knowledge base.

#### re

* `docid`: A `string` feature representing the document identifier.
* `tokens`: A `list` of `string` features representing the tokens in the document.
* `ner_tags`: A `list` of classification labels, corresponding to the NER task.
* `relations`: A `list` of `struct` features containing:
   * `id`: a `string` feature representing the relation identifier.
   * `head_start`: token offset start of the head entity, an `int32` feature.
   * `head_end`: token offset end of the head entity, an `int32` feature.
   * `head_type`: a classification label for the head entity type.
   * `tail_start`: token offset start of the tail entity, an `int32` feature.
   * `tail_end`: token offset end of the tail entity, an `int32` feature.
   * `tail_type`: a classification label for the tail entity type.
   * `type`: a classification label for the relation type. The mapping from ID to relation type is as follows:

```json
{
  "ts:executes": 0,
  "org:develops_or_provides": 1,
  "ts:contains": 2,
  "ts:made_of": 3,
  "ts:uses": 4,
  "ts:supports": 5,
  "met:employs": 6,
  "met:processes": 7,
  "mat:transformed_to": 8,
  "org:collaborates": 9,
  "met:creates": 10,
  "met:applied_to": 11,
  "ts:processes": 12
}
```

### Data Splits

Please add information about your data splits here. For example:

  * **train**: X samples
  * **validation**: Y samples
  * **test**: Z samples

## Dataset Creation

The dataset was created by converting JSON files exported from the Inception annotation tool. The `inception_converter.py` script was used to process these files. This script uses the `dkpro-cassis` library to load the UIMA CAS JSON data and `spaCy` for tokenization and creating BIO tags for the NER task. The data was then split into three separate files for NER, EL, and RE tasks.

## Considerations for Using the Data

### Social Impact of Dataset

More Information Needed

### Discussion of Biases

More Information Needed

### Other Known Limitations

More Information Needed

## Additional Information

### Dataset Curators

Amir Safari

### Licensing Information

Please specify the license for this dataset.

### Citation Information

Please provide a BibTeX citation for your dataset.

```bibtex
  author    = {Amir Safari},
  title     = {Text2Tech Curated Documents},
  year      = {2025},
  publisher = {Hugging Face}
}
```