David Thau, Shawn Bowers, Bertram Ludäscher
The CleanTax framework relates (aligns) taxonomies (inclusion hierarchies) to one another using relations drawn from the RCC-5 algebra. The taxonomies, represented as partial orders with additional constraints, can frequently (but not always) be represented with RCC-5 relations as well. Given two aligned taxonomies, CleanTax can infer new relations (articulations) between their concepts, detect inconsis- tent alignments, and merge taxonomies. Inference and inconsistency detection can be performed by a variety of reasoners, and in cases where all relations can be described by the RCC-5 algebra, qualitative spatial reasoners may be applied. When inferring new articulations between taxonomies, CleanTax often poses many highly related queries of the nature "given what we know about the relations between two taxonomies, T1 and T2 , what do we know about the relationship between concept A in T1 and concept B in T2 ?" This context of posing many (millions) of simple, but highly related queries motivates the need for qualitative reasoning systems that can per- form batch jobs and leverage reasoning performed in the past to optimize answering queries about similar situations. This paper describes the CleanTax framework and argues for the development of benchmarks that take throughput into consideration, as well as single-query response time.