Building Classes in Object-Based Languages by Automatic Clustering

The paper deals with clustering of objects described both by properties and relations. Relational attributes may make object descriptions recursively depend on themselves so that attribute values cannot be compared before objects themselves are. An approach to clustering is presented whose core element is an object dissimilarity measure. All sorts of object attributes are compared in a uniform manner with possible exploration of the existing taxonomic knowledge. Dissimilarity values for mutually dependent object couples are computed as solutions of a system of linear equations. An example of building classes on objects with self-references demonstrates the advantages of the suggested approach.