Techniques are provided for predicting the usage of a document collection
given proximal cue information in the documents, a starting point and the user's
information needs. A document collection topology matrix is created indicating
links between document content portions. The link entry documents are analyzed
for proximal cue words based on link URL, surrounding text and title. For image
links, the connected to document information may also be used. Proximal cue words
are added to a matrix relating proximal cue words and links. The proximal scent
matrix indicates a similarity between the user's information need and the proximal
cue word matrix. A distal scent information matrix is also calculated using distal
document information and combined with the proximal scent matrix. Spreading activation
is then applied to the resulting matrix using the starting location for a requested
number of iterations and resulting in a predicted usage of the document collection.
