MUAD-Net: A Multi-scale URL Anomaly Detection Network for Malicious URL Identification

Abstract

With the continuous evolution of cyber attack techniques, the propagation of malicious links has become a major threat in cybersecurity. Attackers induce users to visit target resources by constructing malicious URLs, leading to information theft and malware distribution. Effectively identifying malicious URLs in complex network environments has become an important research issue. To address this problem, this paper adopts an attack life-cycle perspective and proposes a deep learning-based malicious URL detection method. Considering the complex structure, short lifecycle, and fast generation speed of malicious URLs, the method first normalizes URL strings and encodes them as character sequences, then maps characters into vector representations through an embedding layer. A multi-scale convolutional structure is used to extract local pattern features from URL character sequences, combined with a sequence modeling mechanism to capture contextual dependencies, achieving automatic learning of malicious URL features. Experimental results show that the proposed method achieves high detection performance and good generalization ability. Through multi-scale feature extraction and model optimization, this paper effectively improves malicious URL detection performance, demonstrating good application value in real-world network scenarios.