Unsupervised clustering of symbol strings

The Symbol String Clustering Map (SCM) is introduced as a very simple but effective algorithm for clustering strings of symbols in an unsupervised manner. The clustering is based on an iterative learning of the input data symbol strings. The learning uses the principle of winner take all (WTA) and hence requires a similarity measure between symbol strings. A novel and efficient, average based, similarity measure is defined. Unsupervised generation of the data cluster structure results from the use of a lateral inhibition function applied to the update of adjacent nodes on the SCM lattice. A simple coding method to convert time sequences of symbols to simple symbol strings for use in the SCM is described. The SCM is shown to generate clusters for symbol string data sets.