The fractal nature of relevance: A hypothesis

Abstract

This article proposes a new model, based on fractal geometry, for clusters of relevant documents. It reflects the relatively simple iterative search process used by interactive online searchers. The untested model has the additional attractive features of high lighting the logarithmic growth of clusters, which produces complexities in relevance judgments and document clusters not realized by typical models. It indicates that clusters formed using dynamic search strategies appear topologically distinct, indecomposable, and result from chaotic processes. The model also provides an intuitive definition and representation of cluster dimension which differentiates, where typical models do not, between them. The fractal model, then, gives an indication of what I believe are the limits on clustering relevant documents. © 1994 John Wiley & Sons, Inc.