A Hybrid SOM-Based Document Organization System

This paper presents and evaluates a hybrid system to self-organization of massive document collections based on Self-Organizing Maps. The hybrid system uses prototypes generated by a clustering algorithm to training the document maps, thus reducing the training time of large maps. We test the system with two clustering algorithms: k-means and the AY method. The experiments were carried out with the Reuters- 21758 v1.0 collection. The performance of the system was measured in terms of text categorization effectiveness on test set and training time. The experimental results show that proposed system generate pretty good document maps and that the system had similar effectiveness performance with both clustering methods, however the use of k-means generated the smallest training time.