Exploratory database analysis via self-organization

A new numerical technique, most easily classified as a kind of statistical clustering algorithm, is introduced. This technique, based on principles of collective processing and self-organization, supports the dynamic visualization and manipulation of emergent structures present in multidimensional data sets. The algorithm works by first placing database elements randomly in a grid and then aggregating them in a particular way so that statistical regularities are mapped into spatially structured clusters. The grid is displayed to the human analyst who can probe the contents and alter the characteristics of the clusters as they are formed. It is shown how this technique provides a mean of exploration in complex information spaces, such as in a document corpus or in a relational database, effectively supporting an alternative approach to information access based on exploration from multiple perspectives rather than search.