Incognito

Abstract

Users leave a trail of their personal data, interests, and intents while suring or sharing information on the Web. Web data could therefore reveal some private/sensitive information about users based on inference analysis. The possible identiication of information corresponding to a single individual by an inference attack holds true even if the user identiiers are encoded or removed in the Web data. Several works have been done on improving privacy of Web data through obfuscation methods [7,12,18,32]. However, these methods are neither comprehensive, generic to be applicable to any Web data, nor efective against adversarial attacks. To this end, we propose a privacy-aware obfuscation method for Web data addressing these identiied drawbacks of existing methods. We use probabilistic methods to predict privacy risk of Web data that incorporates all key privacy aspects, which are uniqueness, uniformity, and linkability of Web data. The Web data with high predicted risk are then obfuscated by our method to minimize the privacy risk using semantically similar data. Our method is resistant against adversary who has knowledge about the datasets and model learned risk probabilities using diferential privacy-based noise addition. Experimental study conducted on two real Web datasets validates the signiicance and eicacy of our method. Our results indicate that the average privacy risk reaches to 100% with a minimum of 10 sensitive Web entries, while at most 0% privacy risk could be attained with our obfuscation method at the cost of average utility loss of 64.3%.

Aquilis

“…However, such settings may be confusing for novice users, leading them to involuntarily expose themselves to privacy risks. Throughout this paper, we consider Privacy Risk as defined by Masood et al [27]. A user's privacy is at risk when his or her Web data is distinguishable from other users, has little or no diversity, or is linkable to an individual with high confidence based on the user's Personal Identifiable Information (PII).…”

“…We collect human judgment regarding the sensitivity of 1080 input texts (randomly collected from Reddit and AOL Search Log) using Amazon Mechanical Turk (AMT) 6 . We select input texts belonging to the following topics: politics, religious affairs, legal problems, healthy lifestyle, pets, smoking, and cancer, as per the studies of Biega et al [5] and Masood et al [27]. These texts are between 7 and 35 words long with an average of 16.3 words.…”

“…We then develop a labelled privacy dataset of over 1000 input texts that considers the privacy risk relative to the exposure level of three applications. On this dataset, Aquilis significantly outperforms state-of-the-art methods (F1 score 8% higher than Incognito [27], 15% higher than R-sensitivity [5], and 40% higher than Entropy [42]) as it is the first solution that considers the exposure level of the application in the privacy risk. Finally, we conclude with a user experiment on 35 participants with high technological literacy and privacy awareness to evaluate the objective (difference between participants and Aquilis estimation of the privacy risk) and subjective accuracy (how much participants agree with Aquilis' recommendations).…”

“…There have been many studies on improving the privacy of users online through obfuscation methods. Masood et al [27] propose Incognito, a method for Web data risk prediction combined with an obfuscation technique to protect the users' privacy by adding noise to the data. VACCINE [46] proposes a design methodology for data leakage prevention based on the theory of contextual integrity (CI).…”

Measuring and Analysing the Chain of Implicit Trust

“…With the above in mind, we then proceed to inspect if suspicious or even potentially malicious third-parties are loaded via these long dependency chains (Section 4). We do not limit this to just traditional malware, but also include third-parties that are known to mishandle user data and risk privacy leaks [5,8,16,41,43,52]. Example threats include the re-identification of users in the anonymised AOL search histories, the Netflix training data that was attacked, and the Massachusetts hospital discharge data [16,43,52].…”