Reproduce “SiNMULI: Novel Signed Network Approach for Malicious URL Identification” on OpenResearch

About this paper

Avijit Gayen, Sayan Mondal, Angshuman Jana

Researchers at the Indian Institute of Information Technology, Guwahati, developed SiNMULI, a framework that models malicious URL identification as a signed network classification problem, leveraging a novel directed balance theory to infer trust relationships from hyperlink structures. This approach achieved 99.89% accuracy and a remarkably low false-positive rate of 3 × 10⁻⁵, providing an interpretable and robust method for detecting malicious URLs without extensive training.

Key insights

• Hyperlink structures, when modeled as signed networks, offer a robust and interpretable signal for identifying malicious URLs, overcoming limitations of static feature analysis.
• A novel Directed Balance Theory, adapted for asymmetric trust/distrust relationships, effectively infers the polarity of unknown connections and entities in adversarial web environments.
• This network-centric approach provides a lightweight, training-free, and resilient solution against obfuscation, capable of classifying unseen malicious URLs without requiring model retraining.

Summary by alphaXiv.