Generality-based conceptual clustering with probabilistic concepts

Statistical research in clustering has almost universally focused on data s ets described by continuous features and its methods are difficult to apply to tasks involving symbolic features. In addition, these methods are seldo m concerned with helping the user in interpreting the results obtained. Mac hine learning researchers have developed conceptual clustering methods aime d at solving these problems. Following a long term tradition in Al, early c onceptual clustering implementations employed logic as the mechanism of con cept representation. However, logical representations have been criticized for constraining the resulting cluster structures to be described by necess ary and sufficient conditions. An alternative are probabilistic concepts wh ich associate a probability or weight with each property of the concept def inition. In this paper, we propose a symbolic hierarchical clustering model that makes use of probabilistic representations and extends the traditiona l ideas of specificity-generality typically found in machine learning. We p ropose a parameterized measure that allows users to specify both the number of levels and the degree of generality of each level. By providing some fe edback to the user about the balance of the generality of the concepts crea ted at each level and given the intuitive behavior of the user parameter, t he system improves user interaction in the clustering process.