sdgBERT
| Property | Value |
|---|---|
| Base Model | bert-base-uncased |
| License | MIT |
| Language | English |
| Accuracy | 90% |
| Training Data | OSDG Community Dataset |
What is sdgBERT?
sdgBERT is a specialized NLP model designed to classify text according to the United Nations Sustainable Development Goals (SDGs). Developed at Deakin University, this model is fine-tuned from bert-base-uncased and achieves impressive 90% accuracy in SDG classification tasks. The model supports classification for the first 16 UN SDGs, making it a valuable tool for organizations and researchers working on sustainable development initiatives.
Implementation Details
The model was trained using carefully selected hyperparameters: 3 epochs, a learning rate of 5e-5, and a batch size of 16. It leverages the OSDG Community Dataset for training, ensuring broad coverage across various industries and academic research fields. The model achieves a Matthews correlation of 0.89, indicating robust classification performance.
- Fine-tuned from bert-base-uncased architecture
- Trained on diverse, multi-industry dataset
- Optimized for SDG text classification
- Available through Hugging Face's transformers library
Core Capabilities
- Text classification across 16 UN SDGs
- High-accuracy predictions (90%)
- Processing of English language text
- Easy integration with existing NLP pipelines
- Support for both direct text and PDF document analysis
Frequently Asked Questions
Q: What makes this model unique?
sdgBERT is specifically optimized for SDG classification, making it a specialized tool for sustainability-related text analysis. Its high accuracy and broad training dataset make it particularly valuable for organizations working with UN SDG alignment.
Q: What are the recommended use cases?
The model is ideal for analyzing research papers, corporate reports, policy documents, and any text content that needs to be classified according to UN SDGs. It can be used through direct API calls or via provided demo interfaces for both text and PDF analysis.
