Section: New Results

Addressing Middleware Challenges in Large Scale Mobile Social Networks of the Future

Participants : Sara Hachem, Valérie Issarny, Animesh Pathak, Amir Seyedi.

With the increased prevalence of advanced mobile devices (the so-called “smart” phones), interest has grown in Mobile Social Ecosystems (MSE), where users not only access traditional on line Web-based social networks using their mobile devices, but are also able to use the context information provided by these devices to further enrich their interactions. In complex mobile social ecosystems of the future, the heterogeneity of software platforms on constituent nodes, combined with their intrinsic distributed nature and heterogeneity in representation of data and context, as well as user's privacy and trust concerns, raises the need for middleware support for the development of mobile social applications. We believe that the development of mobile social applications can be greatly simplified by the presence of middleware support. To that end, we have been working on addressing the following challenges:

  • Semantic models for mobile social ecosystems. In order to enable re-use of data between different social applications run by the same user, we have proposed an expressive and extensible model using semantic techniques to represent MSE and the interactions possible in them. This supports semantic interoperability between separately developed applications and minimizes resource-consuming operations such as data mapping and replication.

  • Efficient decentralized storage of social data. Instead of storing the social knowledge of the whole world with a single provider — a practice performed today by common social networks such as Facebook — which can lead to privacy issues, our research endeavors to propose a middleware using which users can store their personal knowledge in a distributed manner on the devices owned by them (e.g., smart phone, home desktop, laptop). This also allows users to provide selective access to other users based on semantically defined access control policies.

  • Socially aware policies for access control. Since social data is private and sensitive in nature, we have proposed a policy framework [21] where the user can specify both the data to be protected as well as the relevant set of peers with access to that data in a socially-aware manner (e.g, “only let my colleagues know my location during weekdays from 9 – 5”). This policy framework can be used as a guard around the user's knowledge base, allowing access only to authorized peers. We are also working on providing end-users an easy to use editor so as to be able to specify these socially-aware policies easily.

  • Social data extraction from existing sources. Our research includes work in enabling users to populate their social knowledge base by extracting data from their existing repositories. We have identified two types of sources of such data. The first already contain social links such as “friendship” in addition to general information, while the second do not contain social links, but may contain information which can be correlated to infer social links (e.g., call and SMS logs). We are working on a framework where adapters can be written for the former using their API to import their data; while for the latter, inference algorithms can be used to correlate data and guess/recommend social links.

  • Inferring trust from proximity. In mobile social network, highly sensitive private data is at risk of being shared with unwanted peers, since users may not have any knowledge about the users they socially connect with. Trust management then appears as a promising decision support for mobile users in establishing social links. However, while the literature is rich in trust models, most approaches lack appropriate trust bootstrapping, i.e., the initialization of trust values. In [24] , we address this challenge by introducing proximity-based trust initialization based on the users' behavioral data available from their mobile devices or other types of social interactions. The proposed approach is further assessed in the context of mobile social networking using users behavioral data collected by the MIT reality mining project. Results show that the inferred trust values correlate with the self-reported survey of users relationships.

We have incorporated our research in the above areas into Yarta [25] , a middleware for mobile social applications. Our prototype middleware, as discussed in §  5.6 , currently supports application development for laptops as well as Android-powered smart phones, providing distributed storage of semantically-modeled social knowledge guarded by a rich policy framework.