A Novel Webpage HiddenLink Detection Method Using BERT and Ternary Decision Approach

Webpage hiddenlink is a typical form of cybersecurity threat, whereby malicious actors surreptitiously embed these links with high-authority webpages associated with governmental, educational, or corporate entities. Subsequently, they exploit these webpages to bolster their SEO (Search Engine Optimization) or engage in other illicit activities. Such practices inflict severe security risks and tarnish the reputation of the aforementioned websites. Against this backdrop, this paper presents a web-based hiddenlink detection methodology, founded upon the employment of the Bert model in conjunction with a ternary decision algorithm. The approach involves training the Bert model to discern features pertinent to hiddenlinks in textual content. Leveraging this model, it extracts feature vectors encapsulating textual information from the target webpage. Subsequently, these feature vectors undergo classification using the three-pronged decision algorithm, which, in turn, enables the identification of hiddenlinks. In instances where an unambiguous classification is unattainable, specifically concerning text labels positioned within the algorithm’s boundary domain, the minHashing algorithm is utilized to gauge the similarity of recorded text fields, leading to the establishment and preservation of text label categories. Consequently, any anomalous text in the database is subjected to manual review and scrutiny, thereby updating the dataset and retraining the Bert model. The proposed methodology remarkably enhances the system’s accuracy in detecting hiddenlinks on webpages while simultaneously providing a continuous feedback loop for refining the existing system. As a result, it effectively and comprehensively detects hiddenlinks within websites.