| --- |
| library_name: transformers |
| tags: |
| - phishing-detection |
| - binary-classification |
| - bert |
| - nlp |
| --- |
| |
| # Model Card for Fine-tuned BERT-Base-Uncased on Phishing Site Classification |
|
|
| ## Model Details |
|
|
| ### Model Description |
|
|
| This model is a fine-tuned version of [BERT-Base-Uncased](https://huggingface.co/google-bert/bert-base-uncased) for phishing site classification. The model predicts whether a website is classified as "Safe" or "Not Safe" based on textual input. |
|
|
| - **Developed by:** [shogun-the-great](https://huggingface.co/shogun-the-great) |
| - **Model type:** Binary Classification (Safe vs Not Safe) |
| - **Language(s):** English |
| - **License:** Apache-2.0 (or specify your license) |
| - **Finetuned from model:** `google/bert-base-uncased` |
| - ** ** |
|
|
| ### Model Sources |
|
|
| - **Dataset:** [shawhin/phishing-site-classification](https://huggingface.co/datasets/shawhin/phishing-site-classification) |
|
|
| ## Uses |
|
|
| ### Direct Use |
|
|
| This model can be directly used for phishing detection by classifying text into two categories: "Safe" and "Not Safe." Typical use cases include: |
|
|
| - Integrating with browser extensions for real-time website classification. |
| - Analyzing textual data for phishing indicators. |
|
|
| ### Downstream Use |
|
|
| Users can fine-tune the model further for specific binary classification tasks or for datasets with similar domains. |
|
|
| ### Out-of-Scope Use |
|
|
| This model might not perform well for: |
| - Non-English text. |
| - Adversarial phishing attacks or heavily obfuscated text. |
| - Tasks unrelated to text-based classification. |
|
|
| ## Bias, Risks, and Limitations |
|
|
| ### Bias |
|
|
| The model's predictions are influenced by the dataset used during fine-tuning. If the training data contains biases, these may reflect in the predictions. |
|
|
| ### Risks |
|
|
| - False positives: Legitimate websites flagged as phishing. |
| - False negatives: Some phishing sites might not be detected. |
| - Potential vulnerabilities to adversarial examples. |
|
|
| ### Recommendations |
|
|
| - Regularly update the dataset and model to stay aligned with emerging phishing patterns. |
| - Use in combination with other security measures for robust phishing detection. |
|
|
| ## How to Get Started with the Model |
|
|
| You can load the fine-tuned model directly from the Hugging Face Hub: |
|
|
| ```python |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification |
| |
| # Load the tokenizer and model from Hugging Face Hub |
| model_name = "shogun-the-great/finetuned-bert-phishing-site-classification" |
| tokenizer = AutoTokenizer.from_pretrained(model_name) |
| model = AutoModelForSequenceClassification.from_pretrained(model_name) |
| |
| # Example usage |
| text = "Enter your login credentials to claim a free reward!" |
| inputs = tokenizer(text, return_tensors="pt", truncation=True) |
| outputs = model(**inputs) |
| |
| # Get the predicted label |
| logits = outputs.logits |
| prediction = logits.argmax(dim=-1).item() |
| print("Prediction:", "Not Safe" if prediction == 1 else "Safe") |
