An Approach toward Automated Generation of Contextual Privacy Policies for Mobile Applications

Abstract

Privacy policies are the predominant mechanism for notifying mobile application users about data practices, yet their length and legal language make them difficult to consult. Contextual Privacy Policies (CPPs) address this problem by presenting short policy segments beside the graphical user interface (GUI) elements to which they apply. This paper presents SeePrivacy, a multimodal framework that detects privacy contexts in mobile screenshots, retrieves related policy segments, and presents them in context. To evaluate SeePrivacy, we establish our benchmark dataset, named CPP4APP, which covers more than 1,200 privacy-related contexts along with matched privacy policy segments. SeePrivacy achieves 0.88 precision and 0.90 recall to detect contexts, as well as 0.98 precision and 0.96 recall in extracting corresponding policy segments. A human evaluation shows that 77% of responses agreed that the segment helps explain the detected context. An Android SDK implementation executes the pipeline in 9.15 seconds on average, demonstrating the feasibility of delivering CPPs in mobile applications. The introductory video, examples, and code are available at our website: https://cpp4app.github.io/SeePrivacy.