Visual navigation in a 3D space querying context information to clusters is not certainly the most user-friendly and useful interface to the database.
We need to enable the user to move in the 3D space not only making direct VRML move actions, but also putting context information. So the user can start with large movements in the 3D space simply inputting words that contribute to build a desired context every time more precise, and then, in "desired context world", move by direct move actions.
In order to add this capability to the system, we need to address a cluster and then its x,y,z coordinates using a word as key. The requirement is to have a table "word to clusters" (one to many) that should be built during the clustering process.
Considering the environment in which the query "word to clusters" should be performed in real time for many clients connected, a normal database is not fast enough and is not possible, due to the large dimension, download the table on the clients. The solution is to make ZISC server for this task on-line for clients. We need to add on the learning process the operations described in fig.1 where a different ZISC card(s) can learn the association word-class-x-y-x. This process must be repeated for any clustering layer: this means that when clients ask for x-y-z they must supply the word and the clustering level in which the browser is moving. This concept is explained in the next section.