A stepwise modeling approach for individual media semantics

Abstract

The generation of appropriate media descriptions is one of the key issues when developing suitable media management systems for every-day use. One of the more practical problems is the deduction of knowledge from partly large but varying, ambiguous, or just too specific information sources (metadata, attributes, features, etc.). This paper presents a stepwise knowledge modeling approach based on two self-contained, trainable rule sets. Following the principle of divide-and-conquer, the supposed approach separately tackles the problem of general translation (normalization) of diverse information sources and their individual interpretation and semantics for the user.