Human-Oriented Fuzzy Set Based Explanations of Spatial Concepts
Brendan P. Young, Derek T. Anderson, James M. Keller, Frederick E. Petry, Chris Michael, Blake Ruprecht · 2023
This article explores generating fuzzy explanations for applications involving spatial intelligence. Two approaches are presented that produce an explanation of a spatial concept from a fuzzy attributed relation graph acquired via a human-in-the-loop algorithm. The first method automatically yields a total spatial concept with explicit graphical and linguistic explanations by identifying the medoid, i.e., an explanation is derived directly from an observed example of the concept. This has the benefit of being both consistent and correct across all relations. The second method uses Zadeh's extension principle to produce a parts-based explanation, however this solution is not guaranteed to be a previously known member of the concept. However, it has some advantages over the medoid that will be shown. To illustrate these methods, two examples are presented. Example one shows the generation of a single explanation for a two-object concept, which highlights how each method creates an explanation for a single relation. The second example generates an explanation for a multi-object concept, illustrating how an explanation is created for more complex concepts.