Abstract
Web browser permission prompts are frequently triggered without sufficient context, leading to user annoyance, decision fatigue, and potential exposure risks. Existing methods for managing these prompts often rely on aggregate site statistics or historical user behavior, which can fail to account for the immediate relevance of a request to the current page content. Such systems may also inadvertently promote deceptive practices if users are frequently misled into granting permissions.
To address these limitations, a method is disclosed for generating a contextual relevance score for permission requests using on-device multimodal analysis. When a permission is requested, a snapshot of the current viewport and the Document Object Model (DOM) is captured. Visual features, such as icons or map elements, and textual features, such as button labels and nearby headers, are extracted and processed by an on-device machine learning model. This model outputs a relevance score indicating the necessity of the permission relative to the page’s visible intent. The score is utilized to dynamically determine whether to display a prominent permission dialog or a suppressed, non-intrusive interface. This technology improves the user experience by filtering out irrelevant prompts while maintaining access to necessary site functionality.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Le, Anh Minh; Hemp, Judith Ulrike; Bacis, Enrico; Engedy, Balazs Csaba; Klim, Elias; and Bilogrevic, Igor, "Predicting Permission Prompt Desirability Using On-Device Visual and Textual Content Features", Technical Disclosure Commons, (March 16, 2026)
https://www.tdcommons.org/dpubs_series/9533