---
license: cc-by-4.0
language:
- en
- nl
- de
- fr
- it
- is
- cs
- da
- es
- ca
metrics:
- accuracy
- matthews_correlation
pipeline_tag: text-classification
library_name: keras
---
# Aurora SDG Multi-Label Multi-Class Model

<!-- Provide a quick summary of what the model is/does. -->
This model is able to classify texts related to United Nations sustainable development goals (SDG) in multiple languages.

![image](https://user-images.githubusercontent.com/73560591/216751462-ced482ba-5d8e-48aa-9a48-5557979a35f1.png)
Source: https://sdgs.un.org/goals 

## Model Details

### Model Description

<!-- Provide a longer summary of what this model is. -->

This text classification model was developed by fine-tuning the bert-base-uncased pre-trained model. The training data for this fine-tuned model was sourced from the publicly available OSDG Community Dataset (OSDG-CD) at https://zenodo.org/record/5550238#.ZBulfcJByF4.
This model was made as part of academic research at Deakin University. The goal was to make a transformer-based SDG text classification model that anyone could use. Only the first 16 UN SDGs supported. The primary model details are highlighted below:

- **Model type:** Text classification
- **Language(s) (NLP):** English, Dutch, German, Icelandic, French, Czeck, Italian, Danisch, Spanish, Catalan
- **License:** cc-by-4.0
- **Finetuned from model [optional]:** bert-base-multilingual-uncased

### Model Sources
<!-- Provide the basic links for the model. -->
- **Repository:** option 1: https://huggingface.co/MauriceV2021/AuroraSDGsModel ; option 2 https://doi.org/10.5281/zenodo.7304546 
- **Demo [optional]:** option 1: https://huggingface.co/spaces/MauriceV2021/SDGclassifier ;  option 2: https://aurora-universities.eu/sdg-research/classify/ 


### Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

This is a fine-tuned model and therefore requires no further training.


## How to Get Started with the Model

Use the code here to get started with the model: https://github.com/Aurora-Network-Global/sdgs_many_berts


## Training Data

<!-- This should link to a Data Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
The training data includes text from 1.4 titles and abstracts of academic research papers, labeled with SDG Goals and Targets, according to an initial validated query.

See training data here: https://doi.org/10.5281/zenodo.5205672

### Evaluation of the Training data

- Avg_precision = 0.70
- Avg_recall = 0.15

Data evaluated by 244 domain expert senior researchers.

See evaluation report on the training data here: https://doi.org/10.5281/zenodo.4917107 


## Training Hyperparameters

<!-- 
- Num_epoch = 3
- Learning rate = 5e-5
- Batch size = 16
-->

## Evaluation

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->
- Accuracy = 0.9
- Matthews correlation = 0.89

See evaluation report on the model here: https://doi.org/10.5281/zenodo.5603019

## Citation
Sadick, A.M. (2023). SDG classification with BERT. https://huggingface.co/sadickam/sdg-classification-bert 

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->


<!--## Model Card Contact -->