In moving from measuring social interactions to augmenting them, a critical task is the integration of heterogeneous data sources, such as real-world face-to-face contacts, on-line friendships and shared interests (both explicitly stated or implicitly inferred from metadata). This integration holds the key to exposing the social semantics of the measured interactions, and paves the door to new kinds of applications that assist social networking, support social browsing, and provide recomendation, and serendipitous discovery.

We believe that the best way to learn about the right research questions is to actually design and build a working system that can be deployed at real-life social gatherings. We partnered with the TAGora project, and specifically with the team of the University of Southampton (Harith Alani, Martin Szomszor and Gianluca Correndo) and designed the Live Social Semantics (LSS) experiment. The basic idea is to focus on a conference gathering and establish a contact between: A) the real-world identities and physical-space relations of conference attendees, B) their identities and relations in web-based systems for social networking and collaborative tagging, and C) their identities in semantic representation of knowledge that describe their interests and collaborations (paper co-authorship, for example). The SocioPatterns project provides the social sensing platform needed for A), while the TAGora project provided the profile-mining and semantic integration techniques required to collect, filter and represent B) and C). The movie below illustrates the experiment concept and the user-facing aspects of the system.

Attendees enroll in Live Social Semantics (LSS) by volunteering to wear a RFID-equipped badge that can detect face-fo-face proximity. This is the real-world part of the experiment. The real-world identity of an attendee is then associated with multiple on-line identities of her choice by creating a user profile in the LSS web interface. LSS currentlty supports Facebook, Flickr, Delicious, and Last.FM. Once a profile has been created, the system will gather data from these systems in the background, record on-line friendships and infer interests and shared interests. The system will also mine for co-membership in communities of practice (COP) using semantic web sources such as the RKBExplorer.

The data collected and integrated by LSS are stored as RDF relations in a triple store, and fed back to the conference attendees in a number of ways. There are public real-time visualizations of the ongoing face-to-face contacts (as in previous SocioPatterns experiments), but now annotated using information and profile pictures from on-line social networks and semantic data. Participants can also visit their account page on the LSS web interface and browse the list of persons they have been in contact with during the event, ranked by the measured strengths of the interactions. On top of that, the system provides an interactive web-based visualization that allows users to browse their ego networks across all supported systems, exploring the interplay of face-to-face time, on-line friendships and shared interest. The system also provides simple forms of recommendation, by suggesting the closure of social triangles that span the supported networks: for example, if attendee A has spent face-to-face time with attendee B, the system can point her to the profile of a third attendee C who is a Facebook friend of both A and B, and hasn’t met A yet at the event.

The architecture of the system and some basic usage statistics are reported in a paper that will be presented in the “Semantic Web in Use” track of the International Semantic Web Conference 2009 (ISWC2009). From a research perspective, the collected data enable to relate, quantitatively, how much an on-line friendship between individuals is predictive of their face-to-face time, and how the structure of the on-line and real-world social networks relate to one another. Results about this will be posted here in the coming weeks.