Scholarlink

<p>In today’s digital age, the rapid increase in online activities has heightened web users’ privacy risks, making them more susceptible to various cyber-attacks. Among these, phishing is one of the most widespread threats in cybersecurity. Phishing employs deceitful tactics to lure individuals and organizations into visiting malicious URLs and disclosing sensitive information, such as passwords, credit card details, and personal data. These attacks are carefully crafted to mislead users into believing they are interacting with legitimate websites or online services, to steal account information for malicious purposes. To address the limitations of traditional phishing detection methods—such as their inability to detect zero-day attacks, high rates of false positives and negatives, and the need for frequent updates to list-based approaches (e.g., blacklists and whitelists)—this paper presents an advanced approach for identifying malicious URLs by integrating transformer learning and deep learning techniques. Specifically, Distilled Bidirectional Encoder Representations from Transformers (DistilBERT) extracts features from URLs and captures relevant textual information. A hybrid deep learning model combining convolutional neural network (CNN) and long short-term memory (LSTM) layers is then applied to classify the dataset into malicious and legitimate URLs. The novelty of this method lies in combining DistilBERT with a hybrid deep learning model (CNN-LSTM) to detect malicious URLs. The proposed approach was evaluated on two large public datasets to test its effectiveness under various conditions. Experimental results demonstrate that it performs exceptionally well, achieving the highest accuracy of 98.19%, and outperforming some comparable methods in the literature on similar datasets. This proves that the integration of DistilBERT and the hybrid CNN-LSTM model is successful in detecting malicious URLs. DistilBERT optimizes the feature extraction process, while the CNN-LSTM model leverages the strengths of both models and mitigates the limitations of each, resulting in a more comprehensive approach to classification.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An enhanced mechanism for malicious URL detection using deep learning and DistilBERT-based feature extraction

  • 期刊论文

  • 发表日期：

    2025年1月27日

查看全文
  • Rania Zaimi,
  • Khouloud Safi Eljil,
  • Mohamed Hafidi,
  • Mahnane Lamia,
  • Farid Nait-Abdesselam

摘要

In today’s digital age, the rapid increase in online activities has heightened web users’ privacy risks, making them more susceptible to various cyber-attacks. Among these, phishing is one of the most widespread threats in cybersecurity. Phishing employs deceitful tactics to lure individuals and organizations into visiting malicious URLs and disclosing sensitive information, such as passwords, credit card details, and personal data. These attacks are carefully crafted to mislead users into believing they are interacting with legitimate websites or online services, to steal account information for malicious purposes. To address the limitations of traditional phishing detection methods—such as their inability to detect zero-day attacks, high rates of false positives and negatives, and the need for frequent updates to list-based approaches (e.g., blacklists and whitelists)—this paper presents an advanced approach for identifying malicious URLs by integrating transformer learning and deep learning techniques. Specifically, Distilled Bidirectional Encoder Representations from Transformers (DistilBERT) extracts features from URLs and captures relevant textual information. A hybrid deep learning model combining convolutional neural network (CNN) and long short-term memory (LSTM) layers is then applied to classify the dataset into malicious and legitimate URLs. The novelty of this method lies in combining DistilBERT with a hybrid deep learning model (CNN-LSTM) to detect malicious URLs. The proposed approach was evaluated on two large public datasets to test its effectiveness under various conditions. Experimental results demonstrate that it performs exceptionally well, achieving the highest accuracy of 98.19%, and outperforming some comparable methods in the literature on similar datasets. This proves that the integration of DistilBERT and the hybrid CNN-LSTM model is successful in detecting malicious URLs. DistilBERT optimizes the feature extraction process, while the CNN-LSTM model leverages the strengths of both models and mitigates the limitations of each, resulting in a more comprehensive approach to classification.