Explain to Me: Towards Understanding Privacy Decisions

Abstract

Privacy assistants help users manage their privacy online. Their tasks could vary from detecting privacy violations to recommending sharing actions for content that the user intends to share. Recent work on these tasks are promising and show that privacy assistants can successfully tackle them. However, for such privacy assistants to be employed by users, it is important that these assistants can explain their decisions to users. Accordingly, this paper develops a methodology to create explanations of privacy. The methodology is based on identifying important topics in a domain of interest, providing explanation schemes for decisions, and generating them automatically. We apply our proposed methodology on a real-world privacy data set, which contains images labeled as private or public to explain the labels. We evaluate our approach on a user study that depicts what factors are influential for users to find explanations useful.