Proceedings of the ACM on Interactive Mobile Wearable and Ubiquitous Technologies · 2018 · 48 citations · 47 references
ComputingInformation TechnologyData MiningMachine LearningMobile ComputingData ProtectionMobile Device SecurityMobile PrivacyData Type InferenceMetadataMobile App PrivacyMobile Data CollectionPersonal DataPrecisionSensitive Personal DataUrl Analysis
Many smartphone apps collect potentially sensitive personal data and send it to cloud servers. However, most mobile users have a poor understanding of why their data is being collected. We present MobiPurpose, a novel technique that can take a network request made by an Android app and then classify the data collection purposes, as one step towards making it possible to explain to non-experts the data disclosure contexts. Our purpose inference works by leveraging two observations: 1) developer naming conventions (e.g., URL paths) of ten offer hints as to data collection purposes, and 2) external knowledge, such as app metadata and information about the domain name, are meaningful cues that can be used to infer the behavior of different traffic requests. MobiPurpose parses each traffic request body into key-value pairs, and infers the data type and data collection purpose of each key-value pair using a combination of supervised learning and text pattern bootstrapping. We evaluated MobiPurpose's effectiveness using a dataset cross-labeled by ten human experts. Our results show that MobiPurpose can predict the data collection purpose with an average precision of 84% (among 19 unique categories).
47
William Enck, Peter Gilbert, Seungyeop Han et al. · ACM Transactions on Computer Systems · 2014 · 1.2K citations · Full text
Eugene Agichtein, Luis Gravano · 2000 · 1.2K citations
Adrienne Porter Felt, Elizabeth Ha, Serge Egelman et al. · 2012 · 973 citations
Kathy Wain Yee Au, Yi Zhou, Zhen Huang et al. · 2012 · 641 citations