TECH Signal 505
Scroll-event burstiness and memory coefficients classify scraper bots with 73.4% accuracy in controlled test
↗
A researcher applied burstiness and memory coefficients, originally from complex-systems event analysis, to JavaScript scroll-event timings, training a two-feature LightGBM model that distinguished human sessions from AI browsing agents with 73.4% accuracy on a controlled dataset.
Request-timing analysis alone fails to separate headless-browser bots like ClaudeBot from humans, but scroll behaviour offers a new signal that bots have not yet mimicked. The approach is limited to scrollable pages and the current model confuses ChatGPT Agent with humans, but combined with mouse-movement or typing features it could become a practical anti-scraping layer for content-heavy sites.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Burstiness and memory coefficients calculated from JavaScript scroll events per page produced distinct distributions for humans versus AI agents, with humans showing higher burstiness and near-zero memory.
A LightGBM model using only those two features achieved 73.4% classification accuracy, with most errors coming from confusion between humans and ChatGPT Agent.
The method is ineffective on pages that do not require scrolling and is most promising for blogging or information sites like Wikipedia.
THE CLUSTER