As such, the adversary will have stronger adversarial capabilities if the dTest function is accessible to the adversary as an oracle; and, therefore, based on this characterization, we define two security games denoted as weak and strong notions, respectively.
For strong adversaries, location questions had the highest confidence rating (i.e., strong adversarial users were more confident when guessing location questions compared to other six question types) followed by activity questions.
Intuitively, since a strong adversary has significant knowledge regarding a user's schedule (e.g., girlfriend), strong adversarial users are more likely to gain access to the system by answering questions more accurately compared to naive adversarial users.
"It really cuts across the strong adversarial character of Australian politics.
As we expected, strong adversarial users performed better than naive adversarial users, since strong adversarial users have significant knowledge regarding legitimate users compared to naive adversarial user.
When it comes to adversarial users, strong adversarial users appeared to be generally neutral (median: 3) in their confidence ratings.
Asking higher weight questions significantly reduced the accuracy of strong adversarial users, whereas legitimate users' accuracy is not affected significantly.
For evaluation, to simulate strong adversarial users, we recruited volunteers in pairs (e.g., close friends, married couples).
While the effect of zoom level on accuracy is found to be significant for legitimate and strong adversarial users, it is insignificant for naive adversarial users.
Out of 7672 responses, 2865 responses were from legitimate users, 2553 responses were from strong adversarial users, and 2254 responses were from naive adversarial users.
