Auto-Refreshed Advertisement Detection From Display-Time Distributions

The system implements feedback by monitoring actual user interactions with advertisements and comparing them against expected interaction patterns. When an advertisement is automatically refreshed, the system checks whether the user actually viewed or interacted with the new ad. This feedback mechanism allows the system to distinguish between useful ad refreshes (where users engage) and wasteful refreshes (where users don't interact), enabling budget optimization by reducing spending on ads that are refreshed but never viewed.

The system segments advertisement delivery into distinct categories: user-initiated ad loads and auto-refreshed ad loads. By segmenting the data this way, the system can track and measure each type separately. This segmentation is achieved by monitoring the timing and context of ad loads against known auto-refresh schedules, allowing precise measurement of which ads are delivered automatically versus through user action, thereby resolving the measurement precision issue while maintaining delivery efficiency.

The system introduces an intermediary layer between the browser and the ad server that mediates the detection process. Instead of directly inspecting obfuscated source code, the intermediary monitors high-level events such as ad load timing, user interaction patterns, and page state changes. This intermediary approach simplifies the detection process by working with observable behaviors rather than trying to parse complex, obfuscated code, thereby maintaining high detection accuracy while reducing the complexity of implementation.