README.md

---
dataset_info:
  features:
  - name: title
    dtype: string
  - name: abstract
    dtype: string
  - name: doctype
    dtype: string
  - name: cluster
    sequence:
      sequence: string
  - name: intervention_area
    sequence:
      sequence: string
  - name: id
    dtype: string
  splits:
  - name: train
    num_bytes: 12305011
    num_examples: 7647
  - name: test
    num_bytes: 426545
    num_examples: 229
  download_size: 7015556
  dataset_size: 12731556
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
---

# ESPON Annotated Scientific Publications Dataset

This dataset consists of annotated scientific publications with the following fields:

- `id`: Unique identifier for the record  
- `title`: Title of the publication  
- `abstract`: Abstract of the publication  
- `cluster_responses`: List of cluster annotations (may be empty in test)  
- `area_responses`: List of area of intervention annotations (may be empty in test)

---

## Splits

- **Train:** Contains full annotations for all fields (`cluster`, `area`, and possibly others)
- **Test:** Contains labels **only** for `cluster` and `area of intervention`

---

## Category Names

### Cluster Labels (6 unique):

- Civil Security for Society
- Climate, Energy and Mobility
- Culture, Creativity and Inclusive Society
- Digital, Industry and Space
- Food, Bioeconomy, Natural Resources, Agriculture and Environment
- Health

### Area of Intervention Labels (38 unique):

- Advanced Materials
- Advanced computing and big data
- Agriculture, forestry and rural areas
- Artificial intelligence and robotics
- Bio-based innovation systems in the bioeconomy
- Biodiversity and natural resources
- Buildings and industrial facilities in energy transition
- Circular Industries
- Circular systems
- Clean, safe and accessible transport and mobility
- Climate science and solutions
- Communities and cities
- Culture, cultural heritage and creativity
- Cybersecurity
- Democracy and Governance
- Disaster-resilient societies
- Emerging enabling technologies
- Energy storage
- Energy supply
- Energy systems and grids
- Environmental and social health determinants
- Environmental observation
- Food systems
- Healthcare systems
- Health throughout the life course
- Infectious diseases, including poverty-related and neglected diseases
- Industrial competitiveness in transport
- Key digital technologies
- Manufacturing technologies
- Net-zero and less polluting Industries
- Next generation internet
- Non-communicable and rare diseases
- Protection and security
- Seas, oceans and inland waters
- Smart mobility
- Social and economic transformations
- Space, including Earth observation
- Tools, technologies and digital solutions for health and care, including personalised medicine

---

## Annotation Methodology

The training data was annotated using an **ensemble of large language models**.  
Specifically:

- Initial predictions were generated independently by:
  - `Meta-Llama-3.1-8B-Instruct`
  - `deepseek-ai/DeepSeek-R1-Distill-Llama-70B`
- For each item, the final label was determined by **majority vote** among these models.
- In cases where there was no majority (i.e., a tie), a **tie-breaker model** (`microsoft/WizardLM-2-8x22B`) was used to select the final label.

This approach aims to increase annotation reliability by leveraging multiple state-of-the-art models and resolving uncertainty with a strong tie-breaker model.



## Number of Samples

| Split  | Number of samples |
|--------|------------------|
| Train  | 7,647            |
| Test   | 229              |

---

## Usage

```python
from datasets import load_dataset

dataset = load_dataset('nicolauduran45/horizon_clusters_annotated')
```