Uncovering Hierarchical Structure in Data Using the Growing Hierarchical Self-Organizing Map

Discovering the inherent structure in data has become one of the major<br> challenges in data mining applications. It requires stable and adaptive<br> models that are capable of handling the typically very high-dimensional<br> feature spaces. In particular, the representation of hierarchical relations<br> and intuitively visible cluster boundaries are essential for a wide range<br> of data mining applications. Current approaches based on neural networks<br> hardly fulfill these requirements within a single model.<br> In this paper we present the Growing Hierarchical Self-Organizing Map<br> (GHSOM), a neural network model based on the self-organizing map. The main<br> feature of this novel architecture is its capability of growing both in<br> terms of map size as well as in a three-dimensional tree-structure in order<br> to represent the hierarchical structure present in a data collection during<br> an unsupervised training process. This capability, combined with the<br> stability of the self-organizing map for high-dimensional feature space<br> representation, makes it an ideal tool for data analysis and exploration.<br> We demonstrate the potential of the GHSOM with an application from the<br> information retrieval domain, which is prototypical both of the<br> high-dimensional feature spaces frequently encountered in today's<br> applications as well as of the hierarchical nature of data.<br> <br> <br> Keywords: self-organizing map (SOM), unsupervised hierarchical clustering, <br> document classification, data mining, exploratory data analysis<br> <br> <br>