Tracking and Visualization of Cluster Dynamics by Sequence-based SOM
Kenichi Fukui, Kazumi Saito, Masahiro Kimura, Masayuki Numao · InTech eBooks · 2010
In this chapter, we introduced the extension of the SOM to visualize the change of dynamic clusters, movement, range, and merger/separation. The sequence weight function was introduced to the neuron topology in order to introduce temporal order to the topology. Using the proposed Sequence-based SOM, an experiment with news articles revealed transition of topics, level of interest, derivation, and diversification/convergence. In addition, class labels were used to obtain weights in order to display the SbSOM results. We applied this display method to time series of medical data. However, further study is needed about overfitting against the training set.