BERT-based Classification Model for AI Generated Text Detection
Model Overview
This BERT-based model is fine-tuned for the task of Ai generated text detection, especially in a TEXT-SQL senario. Please be noted that this model is still in testing phase, its validity has not been fully tested.
Model Details
- Architecture: BERT (bert-base-uncased)
- Training Data: The model was trained on a dataset of 2000 labeled human and ai created questions.
- Training Procedure:
- Epochs: 10
- Batch Size: 16
- Learning Rate: 2e-5
- Warmup Steps: 500
- Weight Decay: 0.01
- Model Performance:
- Accuracy: 85.7%
- Precision: 82.4%
- Recall: 91%
- F1 Score: 86.5%
Limitations and Ethical Considerations
Limitations
The model may not perform well on text that are significantly different from the training data.
Ethical Considerations
Be aware of potential biases in the training data that could affect the model's predictions. Ensure that the model is used in a fair and unbiased manner.
References
- BERT Paper: Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.
- Dataset: Link to the dataset