부산대학교 도서관

Data privacy deals with the sensitive information of individuals and has become a major topic in modern society. Although the practicability of mobile apps has become an essential part of the routines of many people, consumers are increasingly concerned about their data privacy. Most of the time, these privacy policies that are used to collect and share data are lengthy and complex to understand. Therefore, a common user finds it hard to understand before agreeing to the privacy policy terms. This study aims to propose an approach to convey privacy policies in a way that a common user can comprehend. In this paper, we present 10 categories to classify headers and sections of privacy policies, selected after considering both the users’ and domain experts’ views, and an automated privacy policy classification model. Bert, SVM, Naive Bayes, and BiLSTM models are used as the baseline classification models. Our best performing model Bert shows an F1-score of 81%. A dataset of 11K headers and sections of privacy policies classified under the 10 categories and high-level architecture of the privacy policy question answering model are also presented here. Thus, the proposed solution makes awareness and gives an insight to mobile app users on their data privacy.