Issue No. 02 - March/April (2004 vol. 19)
Chihli Hung , University of Sunderland and De Lin Institute of Technology
Stefan Wermter , University of Sunderland
Peter Smith , University of Sunderland
<p>Document clustering is usually performed under the assumption that classification knowledge is unavailable; document classification, however, uses a classified data set for training. The supervised classification approach often achieves greater accuracy than the unsupervised clustering method. If the corpus of documents offers topical categorization, however, clustering can potentially exploit this domain knowledge by moving from an unsupervised to a partially supervised, guided self-organization. In this case, using a neural guided self-organizing network as a metaclassifier on category information offers the opportunity to exploit the domain knowledge. The authors introduce a novel combination of bottom-up dynamic neural learning with top-down symbolic WordNet processing. They show that the hypernym semantic relationship in WordNet complements the neural model, improving classification accuracy and clustering analysis.</p>
SOM, competitive learning, document clustering, document classification, WordNet, neural networks, self-organizing networks, text clustering
S. Wermter, P. Smith and C. Hung, "Hybrid Neural Document Clustering Using Guided Self-Organization and WordNet," in IEEE Intelligent Systems, vol. 19, no. , pp. 68-77, 2004.