Visualization and modification of neural network based information filters - The PaSent approach

23.12.2002

Abstract

A personal health advisor system, PaSent, have been under development using neural network based filter technology for filtering. The filtering process is used for ranking documents of importance to the user based on their discharge letters from hospital or other medical documents.

This thesis investigate these information filters in PaSent for visualization and applied modification through such an interface, The information filters are based on a recurrent neural network so appropriate visual model for interactive modification is needed. Other neural network are also subject for new information retrieval models that might give the PaSent system additional benefits.

Experiment of modification through the visual model have shown that performance for the information filtering can be achieved through such actions. The visual modification tool of the information filters have opened up a gate for a new types of experiments and also lead to discoveries of weaknesses of the present system.