On the Similarity of Eagles, Hawks, and Cows: Visualization of Semantic Similarity in Self-Organizing Maps

We describe an extension to the self-organizing map learning rule enabeling a straightforward visual representation of input data similarity in high-dimensional input structures. The general idea of the extension is to mirror the movement of weight vectors during the training process within a two-dimensional (virtual) output space. The result of the extended training algorithm allows intuitive analysis of the similarities inherent in the input data and most important, intuitive recognition of cluster boundaries.