Final Report Summary - SMARTOSN (Smartphones Meet Online Social Networks) Online social networks and smart mobile devices are two recent developments that have transformed the way that we access information and interact with each other. Apart from being extremely successful in their own right, the two technologies are also getting increasingly integrated. The goal of this project is to investigate issues that arise from the interaction of mobile and social networks. It includes two threads: (I) Fundamentals: Datasets and Network Modeling and (II) Applications and Systems. With regards to Thread I (Fundamentals), our goal was to develop network models that can be efficiently applied to different real-world network data sets pertaining to social and mobile networks. A core question is to identify the minimum set of features that need to be included in the model, so as to accurately capture characteristics of real graphs, while maintaining low complexity. We adopted the systematic framework of dk-series for characterizing the properties of a graph using a series of probability distributions specifying all degree correlations within d-sized subgraphs of a given graph G. Increasing values of d capture progressively more properties of G at the cost of higher complexity. Methods for deterministic graph construction and sampling from dk-series have been developed, albeit not always computationally efficient, up to 2K, while only MCMC techniques are known for d>2. We developed and evaluated methods for 2K and 2.5K graph estimation and construction that achieve different tradeoffs between complexity and accuracy. By “2.5K”, we define graphs that match a given 2K (i.e. joint degree distribution) and some notion of clustering (i.e. a given degree-dependent clustering coefficient). This is important for social networks that exhibit high degree of clustering, thus cannot be modeled using a 2K-distribution alone. Our work was the first that can sample large online social networks and then construct synthetic graphs that resemble the original ones, in terms of the metrics defined above, on the order of a few hours. Our methodology can potentially be applied to network data beyond social networks as well, e.g. for modeling call-description-record datasets.With regards to Thread II (Applications and Systems), the project made two contributions. First, we developed a framework for cooperation of mobile devices within proximity of each other. As a specific application, we focused on web-browsing from a smartphone: the cellular bandwidth is often insufficient to provide good web-browsing experience. We have developed a prototype, called BrowserBuddy, which allows an android phone to connect to another nearby android phone through Bluetooth, and jointly utilize the cellular connections of both phones, in order to speed-up the download of a webpage. Second, we collaborated with Telefonica Research, in Barcelona, Spain, to analyze their call description records for the purpose of implementing a ride-sharing system. In particular, we used the records to identify (i) the commuting patterns of individual users and (ii) groups of people with similar trajectories who could share a ride. We showed that ride-sharing among people having neighboring home and work locations, and potentially picking up passengers along the way, can reduce the number of cars in the city of Madrid by as much as 67%, at the expense of a relatively short detour to pick up/drop off passengers (on the order of 0.5-1km). This demonstrates the high potential of ride-sharing in the city of Madrid and motivates the development of technology and incentives for enabling ride-sharing. Parts of the project that have socio-economic impact include the following. First, ride-sharing in congested cities can help individuals save on gasoline and other car-related costs, while at the same time can reduce traffic and pollution in the city. Work in this area has typically focused on technology, usability, security, and legal issues. However, the success of any ride-sharing technology relies on the implicit assumption that human mobility patterns and city layouts exhibit enough route overlap to allow for ride-sharing on the first place. Our methodology (analyzing cellphone call records to identify human mobility patterns and their overlap) can be applied to estimate how much traffic reduction is possible given a particular city’s layout and the locals’ commuting patterns. This can be a useful tool for city officials to make informed decisions before significant resources are invested on technology and policy making related to ride sharing. Second, our cooperative smartphone apps can improve the experience and reduce the cost for the individual users of smartphones, but can also help all players involved in mobile content delivery save resources, and/or reduce cost; this includes cellular carriers and content providers. Third, the theoretical work on online social network models can enhance the understanding of the structure of online social networks and online human behavior. Finally, and beyond the technical contributions, the project contributed to human resources development by training students. The project involved 3 BS and 2MS students at ITU (all of whom continue pursuing research and advanced degrees in their respective topics) and one Ph.D. student and one postdoc visiting and collaborating from UC Irvine.
