Privacy-Aware Personal Data Storage (P-PDS): Learning how to Protect User Privacy from External Applications

Helping Users Managing Context-Based Privacy Preferences

“…More precisely, let CT X new be a new context and CT X sim the most similar context to CT X new . 4 We take the privacy preference associated with CT X sim , denoted as P P sim ctx , and we ask the user if (s)he would adopt P P sim ctx in CT X new as is or would like to modify it. In the latter case, we let the user modify it obtaining a new preference, denoted as P P mod .…”

“…For instance, Nissenbaum et al [1] show that most of the privacy preference models fail to protect against violations of user privacy preferences because they do not keep into account contextual information. As a matter of fact, many of the existing privacy preferences frameworks (e.g., [2], [3], [4]) do not consider individuals' contextual information to make privacy aware decisions.…”

“…In particular, we compared the proposed approach with a naive approach, that is, a classifier suggesting context-based privacy preferences without leveraging on previously specified context-based privacy preferences. At this purpose, we have extended the approach in [4], where a learning approach has been proposed to suggest non context-based user privacy preferences. The approach in [4] first creates a training dataset of user labels on a set of service requests (e.g., a label is the accept/deny decision on a given service request), then generate a classifier on it, able to automatically decide if a new service request has to be accepted or denied.…”

“…At this purpose, we have extended the approach in [4], where a learning approach has been proposed to suggest non context-based user privacy preferences. The approach in [4] first creates a training dataset of user labels on a set of service requests (e.g., a label is the accept/deny decision on a given service request), then generate a classifier on it, able to automatically decide if a new service request has to be accepted or denied. 1 In order to compare the proposed approach with [4], we extended the latter such as to collect users' labels on service requests complemented with context information.…”

“…The approach in [4] first creates a training dataset of user labels on a set of service requests (e.g., a label is the accept/deny decision on a given service request), then generate a classifier on it, able to automatically decide if a new service request has to be accepted or denied. 1 In order to compare the proposed approach with [4], we extended the latter such as to collect users' labels on service requests complemented with context information. Moreover, we test different supervised machine learning algorithms, namely, Logistic Regression, Random Forest, and Naive Bayesian [5].…”