The Role of Classification in Knowledge Representation and Discovery - 1

Library Trends, Summer, 1999 by Barbara H. Kwasnik

There are many approaches to the process of classification and to the construction of the foundation of classification schemes. Each kind of classification process has different goals, and each type of classification scheme has different structural properties as well as different strengths and weaknesses in terms of knowledge representation and knowledge discovery. The following is a representative sample of some common approaches and structures.

HIERARCHIES

We have inherited our understanding of hierarchical classifications from Aristotle (Ackrill, 1963), who posited that all nature comprised a unified whole. The whole could be subdivided, like a chicken leg at the joint, into "natural" classes, and each class further into subclasses, and so on--this process following an orderly and systematic set of rules of association and distinction. How do we know what a natural dividing place is, and how do we arrive at the rules for division and subdivision? According to Aristotle, only exhaustive observation can reveal each entity's true (essential) attributes, and only philosophy can guide us in determining the necessary and sufficient attributes for membership in any given class. In fact, according to Aristotle's philosophy, it is only when an entity is properly classed, and its essential properties identified, that we can say we truly know it. This is the aim of science, he claims--i.e., to unambiguously classify all phenomena by their essential (true) qualities.

While Aristotle's legacy is alive in spirit in modern applications of classification, most practitioners recognize that a pure and complete hierarchy is essentially possible only in the ideal. Nevertheless, in knowledge domains that have theoretical foundations (such as germ theory in medicine and the theory of evolution in biology), hierarchies are the preferred structures for knowledge representation (see, for example, the excerpt from the Medical Subject Headings [MeSH] in Figure 1).

Figure 1 Hierarchy: Excerpt from MeSH (Medical Subject Headings).(2)

EYE DISEASES
      CONJUNCTIVAL DISEASES
             CONJUNCTIVAL NEOPLASM
             CONJUNCTIVITIS
                    CONJUNCTIVITIS, ALLERGIC
                    CONJUNCTIVITIS, BACTERIAL
                                    OPHTHALMA NEONATORUM
                                    TRACHOMA
                    CONJUNCTIVITIS, VIRAL
                    KERATOCONJUNCTIVITIS
                    REITER'S DISEASE
      CORNEAL DISEASES
             ETC.

Based on the MeSH excerpt in Figure 1, note that hierarchies have strict structural requirements:

* Inclusiveness. The top class (in this case, EYE DISEASES) is the most inclusive class and describes the domain of the classification. The top class includes all its subclasses and sub-subclasses. Put another way, all the classes in the example are included in the top class: EYE DISEASES.

* Species/differentia. A true hierarchy has only one type of relationship between its super- and subclasses and this is the generic relationship, also known as species/differentia, or more colloquially as the is-a relationship. In a generic relationship, ALLERGIC CONJUNCTIVITIS is a kind of CONJUNCTIVITIS, which in turn is a kind of CONJUNCTIVAL DISEASE, which in turn is a kind of EVE DISEASE.