Reproduce “URLNet: Learning a URL Representation with Deep Learning for Malicious URL Detection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Hung Le, Quang Pham, Doyen Sahoo, Steven C.H. Hoi
URLNet presents an end-to-end deep learning framework that learns non-linear URL embeddings through joint character-level and word-level Convolutional Neural Networks for malicious URL detection. The framework attained an AUC of 0.9929 on a 15-million URL dataset, outperforming traditional machine learning baselines and substantially reducing the need for manual feature engineering.
Key insights
- Deep learning, specifically a dual-branch CNN architecture, can automatically extract compact and effective URL feature representations, eliminating the need for manual feature engineering.
- Combining character-level and word-level information provides complementary insights, allowing the model to capture both fine-grained structural patterns and broader semantic context within URLs.
- Novel word embedding strategies, such as character-level word embeddings for individual tokens, are crucial for robustly representing rare and out-of-vocabulary words prevalent in dynamic URL datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.