Semantic computing of simplicity in attributed generalized trees

Abstract

In earlier work, Attributed Generalized Tree (AGT) structures, having vertex labels, edge labels, and edge weights have been introduced. AGTs can represent knowledge in domains containing rich semantic/pragmatic object-centered descriptions as well as complex relations between objects. Therefore, AGTs have applications in many domains such as health, business, and finance (e.g., insurance underwriting). In this paper, we introduce a function to quantify the simplicity of an arbitrary AGT. Our simplicity function takes into account branch, position, and weight factors; it maps the structure to a value in the interval [0,1]. The recursive simplicity algorithm performs a top-down traversal of the AGT and computes its simplicity bottom-up. Characteristic properties of the AGT simplicity measure are analyzed, and AGTs in a test dataset are ranked based on their simplicity values computed using our simplicity algorithm. The experimental analysis confirms our expectation that the simplicity value decreases with increasing the complexity of AGT structure.