III-COR-Small: Social Integration of Semantic Annotation Networks for Web Applications

Indiana University, Bloomington IN

Investigators

Abstract

Link analysis algorithms leverage hyperlinks created by authors as semantic endorsements between pages, while social bookmarks provide a way to leverage annotations by information consumers as a source of information about pages. This project explores a novel approach that is a synergy of the two: soliciting annotations from users about the content of pages, in a way that implicitly forms networks of relationships between and among resources and tags. These socially generated relationships are then aggregated to build bottom-up, global semantic similarity networks. Algorithms are developed to construct, analyze, and mine these networks in support of search and recommendation applications, exploratory navigation interfaces, resource management utilities, tag spam detection, and incentive games to accelerate the achievement of critical mass. To extrapolate both annotations about content (tags) and semantic relationships (similarity) from single users to the ``wisdom of the crowd,'' the project investigates an information-theoretic model that extracts semantic assessments from information structures that many users are already maintaining, namely the bookmarks and tags they manage on their browsers or online. This entails the design and evaluation of several network-based measures and algorithms, such as similarity, novelty, centrality, and focus. Among the aims of this model are the exploration of the duality between resources (URLs) and concepts (tags or categories) and the integration of social annotation and collaborative filtering. One way to provide users with immediate value is to integrate client-based taxonomies and server- based folksonomies for social bookmark management. Both traditional users of browser bookmarks and social users of online bookmarks can take advantage of the same semantic maps while retaining the convenience of intuitive browser interfaces and centralized storage. Further information about the project can be found at http://GiveALink.org.