Malicious Uniform Resource Locators (URLs) embedded in emails or Twitter posts have been used as weapons for luring susceptible Internet users into executing malicious content leading to compromised systems, scams, and a multitude of cyber-attacks. These attacks can potentially might cause damages ranging from fraud to massive data breaches resulting in huge financial losses. This paper proposes a hybrid deep-learning approach named URLdeepDetect for time-of-click URL analysis and classification to detect malicious URLs. URLdeepDetect analyzes semantic and lexical features of a URL by applying various techniques, including semantic vector models and URL encryption to determine a given URL as either malicious or benign. URLdeepDetect uses supervised and unsupervised mechanisms in the form of LSTM (Long Short-Term Memory) and k-means clustering for URL classification. URLdeepDetect achieves accuracy of 98.3% and 99.7% with LSTM and k-means clustering, respectively.

嵌入电子邮件或Twitter帖子中的恶意统一资源定位符（URL）已被用作诱使易受攻击的Internet用户执行恶意内容的手段，从而导致系统，骗局和大量网络攻击。这些攻击可能会造成损害，从欺诈到大规模数据泄露，造成巨大的财务损失。本文提出了一种名为URLdeepDetect的混合深度学习方法，用于单击时间URL分析和分类以检测恶意URL。URLdeepDetect通过应用各种技术来分析URL的语义和词汇特征，包括语义矢量模型和URL加密，以确定给定的URL是恶意的还是良性的。URLdeepDetect使用LSTM（长期短期记忆）和k均值聚类形式的有监督和无监督机制进行URL分类。通过LSTM和k -means聚类，URLdeepDetect的准确度分别达到98.3％和99.7％。