Authors: Harsha V. Madhyastha (UC Riverside), Megha Maiya
Speaker: Harsha V. Madhyastha
In spite of the widespread use of social networks, the authors contend that obtaining opinions from other people remains hard, particularly from friends and family whose opinions may matter more. The primary goal of this paper is to improve how people internetwork with other people and specifically, to make it easier for people to ask for and give recommendations to each other.
The authors note that users have a number of options to recommend things that they like, since many websites provide “Likes” and “+1”s. And it should be possible for a user to ask a question of his friends and find answers quickly. However, in practice, this is much harder. The authors note that people often do not share their likes and dislikes on the Internet. In the current Push-based paradigm, one can query for recommendations and obtain them with low delay, but the recommendations may not be very complete because of low user participation. On the other hand, people may be much more open to sharing their recommendations with specific people, although this could incur higher delay.
This paper aims to examine the tradeoff between delay and completeness by asking the following question: What if the Push-based paradigm is augmented by a Pull-based paradigm? Users often do not take part in the Push-based paradigm, because it involves effort and also involves privacy concerns. Thus, reducing the effort involved to share information and reducing accidental information leakage should incentivize users to share more information.
In order to achieve this goals, the paper suggests the use of a personal repository, which is a complete log of the users’ actions on the Internet. Subsequently, when a user has a query, a search is performed upon the repositories of all the users’ friends. If any of the friends has relevant information, they can choose if they want to share it. In this way, users can share more with lower effort and also exercise a greater degree of control over their privacy.
Q) Pull style may not be appropriate everywhere since it may introduce other sorts of privacy concerns. Maybe Pull is not always the best option?
A) Yes, you are right; the Pull model is not always the best model since consumers are revealing their questions. But since there is the notion of circles in several social networking sites, perhaps we can leverage circles to provide some extra anonymity. The social network could pass on queries from a Circle, and not from a specific user.
Q) The system uses increased logging. What kind of threat models are there if you log even more?
A) Great question, this is preliminary work. One can imagine encrypting all the data with their own key, but even then, there is no doubt that information leakage, if it occurs, will be much bigger. Another question that comes up is if we are we aiding law enforcement significantly, since we may be providing them much more information about ourselves by signing up for the comprehensive personal repository. However, at this point, the aim of the paper is more to bridge the gap between what is being shared and what is not being shared, and the question of threat models will be addressed more carefully in the future.