| Property | Value |
|---|---|
| Task Type | Question Answering |
| Language | Arabic |
| Training Data | Arabic-SQuADv2.0 |
| F1 Score | 71.49% |
| Exact Match | 65.12% |
What is AraElectra-Arabic-SQuADv2-QA?
AraElectra-Arabic-SQuADv2-QA is a specialized Arabic language model fine-tuned for extractive question answering tasks. Built on the AraElectra architecture, this model has been specifically trained to handle both answerable and unanswerable questions using the Arabic-SQuADv2.0 dataset. The model works in conjunction with a classifier to determine whether questions can be answered from the given context.
Implementation Details
The model was trained using a Tesla K80 GPU with carefully selected hyperparameters including a batch size of 8, learning rate of 3e-5, and utilizing the AdamW optimizer over 4 epochs. It implements dynamic padding and leverages the AraElectra base discriminator architecture.
- Supports both answerable and unanswerable questions
- Includes a separate classifier model for question answerability
- Uses ArabertPreprocessor for optimal text preprocessing
- Achieves 75.95% accuracy on handling unanswerable questions
Core Capabilities
- Extractive question answering in Arabic
- Context-based answer extraction
- Question answerability classification
- Integration with Arabic Wikipedia and custom contexts
Frequently Asked Questions
Q: What makes this model unique?
This model uniquely combines question answering capabilities with question answerability classification, specifically designed for Arabic language understanding. It achieves strong performance on both answerable (67.80% F1) and unanswerable (75.95% accuracy) questions.
Q: What are the recommended use cases?
The model is ideal for Arabic information extraction systems, automated question answering services, and educational tools. It's particularly effective when integrated with Arabic Wikipedia or custom knowledge bases requiring precise answer extraction.