RAVEN Project Description

RAVEN combines storage and distributed file-level intelligence data with application-level data from enterprise portals to build a comprehensive semantic repository of an organization’s information assets. It will add a social layer by enabling users to release non-confidential interpersonal messages to this repository, and by extracting Microformats, RDFa and RDF annotations from retrieved documents. Third-party resources will be added through a Web mining and media monitoring platform, which will be augmented by fine-grained incremental mirroring capabilities.

RAVEN will then determine trends in the frequency and semantic orientation of tokens, terms and concepts within this composite information space built from endogenous and exogenous sources. Ambiguity and subtle incremental change of tonal expressions between different versions of a document complicate sentiment detection and often prevent promising algorithms from being incorporated into commercial applications. RAVEN will use a spreading-activation based approach to address these concerns, increase reliability and make sure that every text fragment can be tagged irrespective of its length. The resulting annotations can then be used to control data flows or backup priorities within an organization, for example, or to track media coverage on a recently launched product line. Animated visualizations of temporal-semantic relations will help analysts comprehend and utilize very large data sets comprising millions of documents. To accommodate additional annotations in the visual representation, an AJAX framework will embed temporal controls into ensembles of tightly coupled views in multiple-screen setups. RAVEN will use this framework investigate the benefits of spherical geometries for user interface design, and provide add-on components to use virtual globes as generic image rendering engines for various types of data.

Deploying these computationally expensive algorithms requires a distributed architecture built upon independent services for large clusters of commodity hardware. To build semantic repositories and let users visually explore their temporal-semantic relations in real time, RAVEN needs to address scalability issues on the algorithmic level, as well as on the system and application levels.