Incorporating asymmetry into SOM and Sammon algorithms for visual map generation

The interest in developing visual word maps that help users to find information in the web has grown. Neural based techniques have been proposed to generate word maps but most of them rely on the use of symmetric measures. We propose extended versions of self organizing maps (SOM) and Sammon mapping algorithms that can handle efficiently asymmetric dissimilarities. The modified algorithms are compared to their symmetric counterparts in a real document database using two objective error measures. Our proposed asymmetric algorithms improve the maps generated by their symmetric counterparts.