Overview
Published in the Scientific and Technical Journal of Information Technologies, Mechanics and Optics, this paper investigates technology identification when static signatures are obscured or affected by dynamic loading. Rand Deeb and Alisa A. Vorobeva combine weak supervision with behavioral observations.
Agreement between Wappalyzer and BuiltWith supplies positive training labels. An automated browser collects features from HTTP headers, cookies, network requests and document structure. A multi-label Random Forest predicts technologies, and association rules learned from training data adjust those predictions.
The evaluation covers 8,594 websites and 122 technologies. On the fixed test split, micro-averaged F1 rose from 0.750 for the Random Forest to 0.763 after rule-based post-processing. These are dataset-specific results. The approach complements signature-based identification and does not establish reliable detection of every hidden or previously unseen technology.
Key facts
- Publication
- 2026
- Published in
- Scientific and Technical Journal of Information Technologies, Mechanics and Optics
- Topics
- Web technology identification · Weak supervision · Machine learning · Association rules