On Privacy of Social Communities in Pervasive Networks (case study)

On Privacy of Social Communities in Pervasive Networks (case study)

Every day, mobile operators collect large amounts users' data, specially the billing, throughput, coverage and usage statistics. The data usually come with explicit information likes duration, cost and location and do great help when we start to analyze people's life patterns. Say, social network analyze (SNA) can do a lot via the data collected by location based social network between mobile users.

    People shares their activities, locations, interests and social networks on their twitter or mindless, which can actually do a great more in analyzing people’s behavior and the way they enjoy life. For example, one may use whatsapp (or wechat) to communicate with his friends and enjoy it. But they may never notice the dynamic relationship change when they make new friends.

    In a research conducted by Nokia, they address this problem of community privacy by taking a comparative analysis of the exposure of social relationships and encounters in a deployed wireless peer-to-peer(P2P) network. Over a four month trial in 2010 with 80 participants, they studied and quantified the extent of leakage of private community information by users, by providing empirical evidence about the network or infrastructure owner’s accuracy of reconstruction of the social communities of people. 

    How they manage their research?

   During four months (March-June 2011), they conducted the trial on the EPFL university campus, in order to collect encounter and proximity data. Similarly to previous data collection campaigns, they programmed and distributed 80 Nokia N900 smartphones to the volunteering participants, sampling a coherent population of master’s students and instructors of two classes taught during the spring semester. The participants were asked to carry their device with them as  frequently as possible, and they were allowed to use it as their primary phone. The complete description of the goals and methods of the questionnaires and interviews is described as follows.

    Why their work is unique? Indeed they are unique in three respects:

    –They provide the first privacy analysis of the extent of exposure of community information in a deployed wireless network.

    –They experimentally evaluate and compare the wireless sniffing stations owner’s accuracy of reconstruction of the social communities of people, based on the observed traffic patterns, with the local proximity and encounter data that is collected by the mobile devices.

    –They characterize the evolution of the social interactions among the participants and evaluate the strength of their interactions by implementing three different social interaction measures that take into account the number, the proximity, the recent and aging effects of social relationships in the under lying wireless network.

   At the end of the trial, they obtained useful information from 66 devices, amounting to almost 10GB of collected log data and over 8 million packets captured by the adversarial network. The procedure of analyze is rather typical, that is, they use the four steps as follows:

    For better analyze the data they obtain，they construct mathematics model to conduct the basic rules. In their trial, they get two sources of proximity information: 

(i) the local device logs collected by the mobile devices and containing encounter (list of neighbors, the time stamps and the RSSI values of received packets)

(ii) the adversarial (sniffing) logs containing the headers of the packets sent by the mobile devices, which include the time stamps and RSSI values of received packets at the sniffing stations, as well as the device ID of the sender.

    The results provide basic evidence about the two distinct levels of community information leakage to external observers, who may be able to infer with high accuracy the different social groups and generic communities of people in pervasive networks, while being much less accurate in determining the affiliation of any particular individual to a community.

    As part of our future work, they intend to pursue the analysis of this dual flow of community information leakage and derive mitigation mechanisms in order to reduce information leakage and the gap between the accuracy of both generic statistics and specific people’s affiliations to communities. They also intend to study the adversary’s accuracy of classification of the communities and their members based on the type of their relationship, such as friends, classmates, study group and strangers.

   What we learn most from their case is that they use a typical close relationship, the students in this school can represent many similar ways, for example, people in the same school, company, workplace and so on. The outcome can be easily applied in similar condition. After all, most of social network is found by the close relationship.  

Reference:

1. Gruteser, M., Hoh, B.: On the anonymity of periodic location samples. Security in Perv. Comp. (2005)

2. Hoh, B., Gruteser, M., Xiong, H., Alrabady, A.: Enhancing security and privacy in traffic-monitoring systems. IEEE Perv. Comp. 5 (2006)

3. Matsuo, Y., Okazaki, N., Izumi, K., Nakamura, Y., Nishimura, T., Hasida, K.:Inferring Long-term User Property based on Users. In: IJCAI (2007)

4. Noulas, A., Musolesi, M., Pontil, M., Mascolo, C.: Inferring interests from mobility and social interactions. In: NIPS Workshop on Analyzing Netw. and Learning w.Graphs (2009)

5.Big Brother Knows Your Friends: Lgor Bilogrevic, Murtuza Jadliwala.Nokia Research Center, Lausanne, Switzerland