How vishwanath-06/fusionphishguard works
FusionPhishGuard is a research framework for classifying URLs and mobile-web links as legitimate or phishing. Its documented workflow tokenizes inputs, extracts lexical, subword, contextual, and LLM representations, fuses and refines features, models URL sequences, and classifies the result. The README documents these stages and branches, but the listed implementation paths under src/ are absent from the supplied tree; component paths are therefore left null except for the two present notebooks. Connections reflect the documented pipeline, not inspected runtime calls.
Main entry points: phishing-with-word-embedding-techniques.ipynb Word2Vec · phishing-with-combinations-w-t-w-llm-t-llm.ipynb MiniLM
Components 14
URL Representations
Feature Fusion
- Gated Attention
- SE Refinement
Classification
- BiLSTM
- Classifier
- Prediction
Other components
- URL Submitter
Connections 19
- URL Submitter → URL Tokenizer: submits URL
README.md - URL Tokenizer → Word2Vec: tokenizes for
README.md - URL Tokenizer → FastText: tokenizes for
README.md - URL Tokenizer → BERT: tokenizes for
README.md - URL Tokenizer → RoBERTa: tokenizes for
README.md - URL Tokenizer → MiniLM: tokenizes for
README.md - URL Tokenizer → Qwen: tokenizes for
README.md - URL Tokenizer → Falcon: tokenizes for
README.md - Word2Vec → Gated Attention: provides features
README.md - FastText → Gated Attention: provides features
README.md - BERT → Gated Attention: provides features
README.md - RoBERTa → Gated Attention: provides features
README.md - MiniLM → Gated Attention: provides features
README.md - Qwen → Gated Attention: provides features
README.md - Falcon → Gated Attention: provides features
README.md - Gated Attention → SE Refinement: fuses into
README.md - SE Refinement → BiLSTM: refines for
README.md - BiLSTM → Classifier: models for
README.md - Classifier → Prediction: classifies as
README.md