User behavior strongly predicts web browsing identity and next sites
Behavioral Fingerprinting and Navigation Prediction in Web Browsing
Machine Learning
Summary
Web browsing might seem random and short, but even small fragments of activity carry strong clues about who is browsing and what they will do next. The authors study how well one can tell users apart from their browsing sessions and predict their next website. They find that users are highly identifiable from short browsing patterns and that future navigation can be forecast well using long-term interaction data combined with recent actions. Using large language models adds little on top of simpler sequence and structure models, showing that repeated user habits dominate prediction.
What this means in practice
- •For security analysts: Detect users from short web session data to identify suspicious behavior in anonymous browsing streams.
- •For web personalization teams: Improve website recommendations by predicting next browsing domains using long-term and recent user interaction data.
Authors
Ralph Elsaghbini, Omran Berjawi, Walid Fahs, Rida Khatoun
Abstract
Web browsing often appears ephemeral: users visit a few websites, complete a task, and move on. However, even short fragments of browsing activity can contain rich and structured behavioral signals. In this work, we conduct a comparative empirical study of two complementary behavioral inference tasks: session-level user identification and next-domain prediction. Both tasks are derived from the same cleaned event stream and evaluated on large-scale anonymous browsing traces, with sessionization and splitting adapted to the temporal requirements of each task. For user identification, we evaluate classical and neural models operating on session-level behavioral and domain features. For next-domain prediction, we combine graph-based modeling with Large Language Models (LLMs). Experimental results show that short browsing sessions are highly identifiable, while future navigation actions are highly predictable from long-term interaction structure combined with recent behavioral context. Furthermore, LLM-derived semantic features yield only marginal gains over purely structural and sequential models, indicating that repeated interaction patterns remain the dominant predictive signal in the evaluated web-browsing setup. These findings highlight the extent to which interaction history substantially contributes to both user identifiability and navigation predictability in browsing traces.