Future Trends

The Self Organizing Map has great potential as a tool for creating useful visualizations of Websites. The present research has begun to develop the techniques necessary to get meaningful and useful results from the SOM neural network. In doing so, a visualization can be created that is as useful as the Web Usage Plot, but more robust in its computation. The SOM technique is relatively fast to compute, and has no restriction on the number of Web pages that can be considered.

However, there is still much more research necessary to successfully use the SOM as part of a professional visualization system. One difficulty in using the SOM is a problem with dealing with sparse data sets. In a small Website sample, it is quite possible that most of the Web pages are never visited, or are visited in one session. This means that of the hundreds of sessions available, a given Web page will have only a single column active. The SOM network will frequently overlook the subtle difference between such pages, in considering the vast similarity in their pattern of not being accessed in so many sessions. Finding a set of parameters and appropriate training regimen for dealing with this problem can be quite time consuming. At present, a variety of different parameters must be experimented with using trial and error in order to find a useful visualization.

Perhaps even more important than the enabling of intuitive visualizations, capturing a Website's usage pattern in a neural network could provide a remarkably versatile component in new Web-based applications. Recommendations could be made to the user on the fly, since when a user goes to a particular Web page, it will be clear which other pages the other users have visited from there. Also, it would be possible to recreate some user behaviors from the network itself, so that novel Website structures can be readily evaluated or compared.