User Adaptive Categorization of Document Collections

Abstract

Methods for the automatic categorization of documents are usually based on a simple analysis of the considered document collection. User specific criteria, e.g. interests in specific topics or keywords, are usually neglected. Therefore, the resulting categorization frequently does not fulfil the user expectancies. In prior work we had developed an approach to cluster document collections by growing self-organizing maps that adapt their structure automatically to the structure and size of the underlying document collection. In this paper, we present an approach to improve the obtained clustering by considering user feedback (in the form of drag-and-drop) to adapt the underlying topology and thus the categorization of documents by the self-organizing map. Furthermore, we briefly present applications for image and text document collections.KeywordsSource NodeWeight VectorDocument CollectionTarget NodeText DocumentThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.