Demo: O’BTW- An Opportunistic, Similarity-based Mobile Recommendation System ∗
Mai ElSherief† ∗, Tamer ElBatt† ∗, Ahmed Zahran† ∗, Ahmed Helmy‡ †
Wireless Intelligent Networks Center (WINC), Nile University, Giza, Egypt ∗Faculty of Engineering, Cairo University, Giza, Egypt ‡Dept. of Computer and Information Science and Engineering, University of Florida, Gainesville, USA
Categories and Subject Descriptors C.2 [COMPUTER-COMMUNICATION NETWORKS]: Miscellaneous; K.4 [ COMPUTERS AND SOCIETY]: Miscellaneous
Keywords Mobile users, Opportunistic apps, temporal, Profiles, Similarity metrics, Proximity based Apps, Matrix vectorization
1.
INTRODUCTION
Motivation and Objective: Earlier social studies, e.g., Homophily [Lazarsfeld and Merton (1954)], have shown that people tend to have similarities with others in close proximity. In our demo, coined O’BTW, we exploit the ubiquity of mobile phones and resource-rich users’ social structure to develop opportunistic similarity-based mobile social networking applications. O’BTW exploits these key social insights to define new peer-to-peer networking regimes. O’BTW creates endless possibilities for highly personalized apps based on phone-to-phone communications, e.g., location-based services, targeted ads, dating and social networking apps and establishing trust. System Overview: O’BTW demo hosts three major research components: I. Introducing generalized mobile user profile structures that capture relevant factors in assessing user similarity. In O’BTW, we employ a general profile that incorporates different facets, like digital (e.g. visited websites) and physical footprints (physically visited sites), ratings, tips, etc. Additionally, our profile incorporates a temporal dimension that enriches the profile and improves the similarity assessment. II. Developing novel similarity assessment tools that feature both high accuracy and low complexity. In O’BTW, we introduce Vectorized Cosine similarity metric for temporal profiles. This metric enjoys the computational simplicity of non-temporal similarity assessment for non-temporal profiles, e.g, Cosine similarity chosen for its superior performance and the richness brought in by incorporating the temporal dimension into the user profile, e.g., SVD-based similarity [1]. III. Designing DTN profile dissemination policies that compromise the tradeoff between forwarding overhead and de∗ This work was funded in part by a Google Faculty Research Award.
Copyright is held by the author/owner(s). MobiSys’13, June 25–28, 2013, Taipei, Taiwan. ACM 978-1-4503-1672-9/13/06.
lay. The optimal policy would maximize the informationtheoretic “knowledge gain” metric (under investigation). Other Apps that exploit the opportunistic encounters are iTrust and Shield. Highlight and Sonar are apps that connect people in proximity and share information, however, these apps reveal the complete identity of the users.
2.
O’BTW DEMO
The O’BTW demo is implemented using the Android SDK. The demo exhibits a “research-focused” use case in which user profiles are built on the spot through a friendly GUI where physical footprints are represented as the technical conferences attended over the past three years. O’BTW first establishes connection between devices in proximity using Bluetooth. Afterwards, it assesses pair-wise, temporal profile similarity using our Vectorized Cosine metric. In case of similarity, O’BTW exchanges the pre-stored tips representing this user’s knowledge base. Privacy is a major issue in profile-based communication systems that we are currently tackling via profile anonymization techniques as a first line of defense and is a major topic of a more comprehensive analysis and investigation. O’BTW is proof-of-concept mobile app by which mobile users can readily, and anonymously, exchange knowledge, ratings and recommendations with others who are “alike” and happen to meet opportunistically. This, in turn, expands our knowledge base from the “legacy” O’BTW limited to “people we know and meet” to the digital O’BTW which leverages “people we encounter and do not know” and ultimately, with DTN forwarding to “people we do not know and have never encountered”.
3.
REFERENCES
[1] W. Hsu, D. Dutta, and A. Helmy. CSI: A paradigm for behavior-oriented profile-cast services in mobile networks. Ad Hoc Networks, 10:1586–1602, 2012.