Search

The rise of cyber threats, particularly URL-based phishing attacks, has tarnished the digital age despite its unparalleled access to information. These attacks often deceive users into disclosing confidential information by redirecting them to fraudulent websites. Existing browser-based methods, predominantly relying on blacklist approaches, have failed to effectively detect phishing attacks. To counteract this issue, we propose a novel system that integrates a deep learning model with a user-centric Chrome browser extension to detect and alert users about potential phishing URLs instantly. Our approach introduces a Knowledge Distilled ELECTRA model for URL detection and achieves remarkable performance metrics of 99.74% accuracy and a 99.43% F1-score on a diverse dataset of 450,176 URLs. Coupled with the browser extension, our system provides real-time feedback, empowering users to make informed decisions about the websites they visit. Additionally, we incorporate a user feedback loop for continuous model enhancement. This work sets a precedent by offering a seamless, robust, and efficient solution to mitigate phishing threats for internet users. [ABSTRACT FROM AUTHOR]

Malicious attacks have developed a prominent risk to the safety of online users, with attackers employing increasingly sophisticated systems to deceive unsuspecting victims. This research focuses on the critical aspect of feature selection in optimizing phishing uniform resource locator (URL) detection system. Feature selection boosts machine learning (ML) and deep learning (DL) by picking vital attributes efficiently. This research paper provides a comprehensive examination of feature selection techniques using five diverse datasets. Various methods, including random forest (RF) select from model, SelectKBest with chi-square statistic, principal component analysis (PCA) and recursive feature elimination (RFE), were employed. The experiments, with a particular emphasis on PCA and fourth dataset, revealed that all four models RF, decision trees (DTs), XGBoost, and multilayer perceptron) achieved 100% accuracy in detecting phishing URL attacks. This underscores the efficacy of feature selection methods in enhancing to a deeper understanding of feature selection's role in bolstering the effectiveness of phishing detection system across diverse datasets, highlighting the importance of leveraging techniques such as PCA for optimal results. [ABSTRACT FROM AUTHOR]

The rise of cyber threats, particularly URL-based phishing attacks, has tarnished the digital age despite its unparalleled access to information. These attacks often deceive users into disclosing confidential information by redirecting them to fraudulent websites. Existing browser-based methods, predominantly relying on blacklist approaches, have failed to effectively detect phishing attacks. To counteract this issue, we propose a novel system that integrates a deep learning model with a user-centric Chrome browser extension to detect and alert users about potential phishing URLs instantly. Our approach introduces a Knowledge Distilled ELECTRA model for URL detection and achieves remarkable performance metrics of 99.74% accuracy and a 99.43% F1-score on a diverse dataset of 450,176 URLs. Coupled with the browser extension, our system provides real-time feedback, empowering users to make informed decisions about the websites they visit. Additionally, we incorporate a user feedback loop for continuous model enhancement. This work sets a precedent by offering a seamless, robust, and efficient solution to mitigate phishing threats for internet users.

The Internet of Things (IoT) has proven its utility across various domains, including healthcare, agriculture, industry, and finance. It comprises Internet-connected devices that offer remote control capabilities. However, this very connectivity exposes these devices to potential cyberattacks. Cybercriminals can exploit the vulnerabilities in these devices by sending deceptive emails or text messages containing malicious links leading to hacker websites or destructive applications. This allows them unauthorized access to connected devices and the acquisition of sensitive personal information. Such malicious tactics are collectively known as phishing and pose one of the most prevalent threats. This article presents an innovative method that harnesses the power of a Deep Neural Network to accurately classify and proactively prevent phishing websites by analyzing their URLs. The method is demonstrated through a smart home use case, aiming to reinforce IoT security and safeguard users' sensitive data by proactively identifying and preventing phishing attacks. By harnessing the power of the Deep Learning model, this innovative technique seeks to enhance online safety and protect users from potential cyber threats. [ABSTRACT FROM AUTHOR]

The current ubiquity of online social networks (OSNs) cannot be overstated, and they have over 4.8 billion users worldwide. These platforms have become integrated into modern life, representing an important means of communication and information sharing. However, this widespread popularity has also drawn the attention of cybercriminals, who seek to exploit OSNs using deceptive Uniform Resource Locators (URLs) as their weapons of choice. Conventional URL-classification methods, which rely on post-access features or static analysis, face significant limitations; they struggle to keep pace with the ever-evolving tactics of cybercriminals, and they often lack the granularity required for precise URL categorization. The methodology proposed herein takes a different path, leveraging the power of an artificial neural network (ANN) in tandem with Bidirectional Encoder Representations from Transformers (BERT) to extract contextual embeddings from URLs. By combining the cutting-edge capabilities of ANNs and BERT, we introduce an efficient approach to safeguarding OSN users from the insidious threats lurking behind deceptive URLs by classifying them into five distinct categories: benign, defacement, phishing, malware, and spam. The proposed approach was found to achieve an impressive accuracy rate of 98.0%, surpassing the previous best of 97.92%. This technique thus has the potential to serve as a crucial defense mechanism for the billions of individuals who rely on OSNs for their social and informational needs. [Display omitted] • A broad exploration of the state of the art of URL detection is conducted. • A context-aware embedding method for detecting fraudulent URLs is introduced. • Traditional binary classification is extended to a multiclass system using ANNs. • A comprehensive dataset encompassing various URL types is curated and used. [ABSTRACT FROM AUTHOR]

Throughout the past few years, phishing attacks have become an increasingly substantial problem for individuals and organizations. In this non-technical attack, the victim is deceived into accessing a malicious URL that downloads a malicious program to access the network or redirects the victim to a page that requests sensitive information. The literature is filled with many research proposals to mitigate this problem however, the dynamic nature and the creativity of the attacker have made it difficult and reoccurring. In this paper, we proposed to use novel deep learning techniques, namely the Temporal convolutional network (TCN) with word embedding, to detect phishing URLs. As a result of our experiments, we found that our model can detect phishing URLs with 98.95% accuracy, 98% precision, 98% recall, and 98% f1-score. This result indicates that our model is effective against phishing attacks. [ABSTRACT FROM AUTHOR]

The learned Bloom filter (LBF) combines a machine learning model (learner) with a traditional Bloom filter to improve the false positive rate (FPR) that can be achieved for a given memory budget. The LBF has recently been generalized by making use of the full spectrum of the learner’s prediction score. However, in all those designs, the machine learning model is fixed. In this letter, for the first time, the design of LBFs is proposed and evaluated by considering the machine learning model as one of the variables in the process. In detail, for a given memory budget, several LBFs are constructed using different machine learning models and the one with the lowest FPR is selected. We demonstrate that our approach can achieve much better performance than existing LBF designs providing reductions of the FPR of up to 90% in some settings. [ABSTRACT FROM AUTHOR]

Deep learning models for phishing URL classification based on character- and word-level URL features achieve the best performance in terms of accuracy. Various improvements have been proposed through deep learning parameters, including the structure and learning strategy. However, the existing deep learning approach shows a degradation in recall according to the nature of a phishing attack that is immediately discarded after being reported. An additional optimization process that can minimize the false negatives by selecting the core features of phishing URLs is a promising avenue of improvement. To search the optimal URL feature set and to fully exploit it, we propose a combined searching and learning strategy that effectively models the URL classifier for recall. By incorporating the deep-learning-based URL classifier with the genetic algorithm to search the optimal feature set that minimizing the false negatives, an optimized classifier that guarantees the best performance was obtained. Extensive experiments on three real-world datasets consisting of 222,541 URLs showed the highest recall among the deep learning models. We demonstrated the superiority of the method by 10-fold cross-validation and confirmed that the recall improved compared to the latest deep learning method. In particular, the accuracy and recall were improved by 4.13%p and 7.07%p, respectively, compared to the convolutional–recurrent neural network in which the feature selection optimization was omitted. [ABSTRACT FROM AUTHOR]

The use of digital devices has increased exponentially in the last decade. Especially, young children spend most of their time surfing for various reasons such as homework. assignments and projects etc. Parental Control is highly important for monitoring the browsing behavior of children. Though several content-based web page classification approaches are available, it requires the entire web page contents for classification purpose. This leads to wastage of bandwidth due to unnecessary downloads. The exponential growth of internet demands URL based classifiers to adapt to the dynamic web, and to make swift decisions on the fly. To address this problem, a deep learning based approach has been proposed in this research that can extract the features only from the URL of a web page. To learn the patterns for Kids-specific web sites automatically, Convolutional Neural Network (CNN) is combined with Bidirectional Gated Recurrent Unit (BGRU) to extract rich context aware features as well as to preserve the sequence information in the URL. By conducting various experiments on the benchmark collection Open Directory Project (ODP), it is shown that an accuracy of 82.04% can be achieved. [ABSTRACT FROM AUTHOR]

In recent years, with the rapid development of internet, telecom operators can gather more and more DPI data. How to mine and classify huge amounts of DPI data in real time is the key point. A DPI data classification method was presented, which combined the URL classification model based on mixed n-gram features and the Doc2Vec based text classification algorithm. The DPI data classification speed and classification accuracy were greatly improved, and good support for precise marketing was provided.

Most Intrusion Detection Systems (IDS) nowadays are signature-based. They are very useful and accurate for detecting known attacks. However, they are inefficient with unknown attacks. Moreover, most of cyber attacks start with malicious URLs. Attackers try to trick users into clicking on malicious URLs. This gives attackers an easy way to launch attacks. To defend against this, companies and organizations use IDS/IPS to detect malicous URLs using blacklist of URLs. This is very efficient with known malicious URLs, but useless with unknown malicious URLs. To overcome this problem, a number of malicious Web site detection systems have been proposed. One of the most promising methods is to apply machine learning detection techniques. In this paper, we present a new lexical approach to classify URLs by using machine learning techniques which patternize the URLs. Our approach is based on natural language processing features which use word vector representation and ngram models on the blacklist word as the main features. Using this technique can help classifier distinguish benign URLs from malicious ones. Our experimentation shows that our approach can achieve a high degree of accuracy at 97.1% in the case of SVM. Moreover, it can maintain a high level of robustness with 0.97 precision and 0.93 recall scores. [ABSTRACT FROM AUTHOR]

Abstract: Web page classification has become a challenging task due to the exponential growth of the World Wide Web. Uniform Resource Locator (URL)‐based web page classification systems play an important role, but high accuracy may not be achievable as URL contains minimal information. Nevertheless, URL‐based classifiers along with rejection framework can be used as a first‐level filter in a multistage classifier, and a costlier feature extraction from contents may be done in later stages. However, noisy and irrelevant features present in URL demand feature selection methods for URL classification. Therefore, we propose a supervised feature selection method by which relevant URL features are identified using statistical methods. We propose a new feature weighting method for a Naive Bayes classifier by embedding the term goodness obtained from the feature selection method. We also propose a rejection framework to the Naive Bayes classifier by using posterior probability for determining the confidence score. The proposed method is evaluated on the Open Directory Project and WebKB data sets. Experimental results show that our method can be an effective first‐level filter. McNemar tests confirm that our approach significantly improves the performance. [ABSTRACT FROM AUTHOR]

A self-learning light-wight (SLW) is proposed.SLW is the first to introduce access relations and have the char-acteristics of feedback and self-learning.SLW approach starts from the seed set which includes known malicious pages.Then,it automatically figures out users with low credibility based on the seed set and the visit relation database.Finally,the access records of these users are used to identify other malicious pages.Experimental results indicate that SLW ap-proach can significantly improve the efficiency of malicious pages detection and reduce the average detection time com-pared with other conventional methods.

Webometrics is gaining big momentum nowadays. Although it is not called by this name in commercial sector, there is a tremendous demand for it. On the low level the majority of webometric studies use a classification. When classification is not defined outside of the webometric study, as for example product portfolio (Feige, 2013), its establishing is tedious human coder work. The process is laborious and result depends on a particular researcher. The patterns in URL can be used in for an automated establishing of classification via tagging, using principle of grounded theory (Thelwall 2013). As such, the patterns in the URL will reduce amount of work in classifying and eliminate the human opinion factor from the process. [ABSTRACT FROM AUTHOR]

Abstract: Unsupervised web page classification refers to the problem of clustering the pages in a web site so that each cluster includes a set of web pages that can be classified using a unique class. The existing proposals to perform web page classification do not fulfill a number of requirements that would make them suitable for enterprise web information integration, namely: to be based on a lightweight crawling, so as to avoid interfering with the normal operation of the web site, to be unsupervised, which avoids the need for a training set of pre-classified pages, or to use features from outside the page to be classified, which avoids having to download it. In this article, we propose CALA, a new automated proposal to generate URL-based web page classifiers. Our proposal builds a number of URL patterns that represent the different classes of pages in a web site, so further pages can be classified by matching their URLs to the patterns. Its salient features are that it fulfills all of the previous requirements, and it has been validated by a number of experiments using real-world, top-visited web sites. Our validation proves that CALA is very effective and efficient in practice. [Copyright &y& Elsevier]