Distributed multimedia document modeling
Luigi Lancieri · 2002
First, the paper describes a model to represent a set of heterogeneous data widely distributed. Secondly, we propose a method to use this model based on a specific learning process. This approach allows us to build up dynamically a "semantic database" consistent with a user's group profile. To reach this goal, the database is built through using data contained in Intranets caches. With approximate criteria of search it becomes possible to retrieve document or multimedia components which best fit with the request. It is also possible to "semantically" compare documents or evaluate the "semantic distance" between a document and a particular theme. The model also provides a method for self-extracting themes from the "semantic database". The first step is to define a semantic architecture based on neural nets. The second step provides an algebraic model of this architecture represented by a single matrix. Doing so it is possible to easily evaluate relations (e.g. semantic links) in multimedia documents, by computing some mathematical properties of this matrix (e.g. Euclidean distances).