Self-organizing concept maps
Masafumi Hagiwara · 2002
Self-organizing concept maps (SOCOMs) based on a neural network model are proposed in this paper. They can arrange concepts or words in a map space using Kohonen's self-organizing map algorithm. One of the most important advantages of the proposed maps is that they employ the idea of k-nearest neighbor (k-NN): they do not require all of the data among concepts or words. The author proposes two kinds of SOCOMs: one is a metric SOCOM, another is a non-metric one. The metric SOCOM uses the information about the metric data such as similarity. The non-metric one uses the information about the rank order of similarity among items. The combination of the idea of k-NN and a non-metric SOCOM is effective to relax the severe requirements on data: it does not require all of the detailed metric information among concepts or words. Computer simulation results have shown the effectiveness of the proposed SOCOM.