Description of the model

In this chapter, we describe a family of integrated categorization and category learning models that process and organize past experience to facilitate responses to future experience. The Cobweb system (Fisher, 1987) and its descendants Classit (Gennari, 1990), Oxbow (Iba, 1991), Labyrinth (Thompson & Langley, 1991), Dædalus (Langley & Allen, 1993), and Twilix (Martin & Billman, 1994) comprise a family of models that share a genealogy, a search strategy, and a heuristic to guide that search. We will often refer to this entire family as Cobweb when the intended meaning is clear from the context.

These systems grew out of machine learning and cognitive science research that explored methods for acquiring concepts in an unsupervised context. In that setting, a teacher does not provide explicit category information for instances as the learner encounters them; instead, the learner must decide how to group or categorize a collection of instances. In contrast to most clustering methods, instances are encountered incrementally; the learner must make appropriate adjustments in response to each one as it comes. The Cobweb family of models view categorization as a conceptualization process or as the formation of ontologies. That is, these models provide answers to the question, ‘How does one form conceptual representations of similar experiences and how might those representations be organized?’ However, these models also address the use of the acquired concepts to handle future situations.