Using Object Concepts to Match Artificial Intelligence Techniques to Problem Types

Abstract

Using object-concepts as a matching framework, we provide guidelines for identifying what types of problems are best served by which knowledge-representation technique. We find that production rules are best for hierarchical classification problems, because they support classification/instantiation of data. Frames are best for data retrieval and inference problems, because, using data abstraction, frames can operate on data within a frame. Finally, semantic networks are best for consequence finding problems, because of independence of the primitives in the hierarchy. Providing guidelines for this matching is important, because the success of different information systems designs have been shown to depend explicitly on problem characteristics.