Web security; Pre-training; Dynamic graph; Temporal data; Anomaly detection

The WPT dataset was specially created for "Web Page Tampering Detection Based on Dynamic Temporal Graph Pre-training" and encompasses over 200,000 regular web pages from 75 websites across the finance, healthcare, and education sectors, in addition to 1,541 tampered examples sourced from zone-h.org. This dataset organizes web pages as nodes and their links as edges within a discrete dynamic graph structure, capturing snapshots at various moments in time. Each node integrates structural, textual, and statistical features into a robust 148-dimensional feature vector for every page.