Self organizing map and sammon mapping for asymmetric proximities

Abstract

Self Organizing Maps (SOM) and Sammon Mapping (SM) are two information visualization techniques widely used in the data mining community. These techniques assume that the similarity matrix for the data set under consideration is symmetric. However there are many interesting problems where asymmetric proximities arise, like text mining problems are. In this work we propose modified versions of SOM and SM to deal with data where the proximity matrix is asymmetric. The algorithms are tested using a real document database, and performance is reported using appropriate measures. As a result, the asymmetric algorithms proposed outperform their symmetric counterparts.