Extending classical reasoning for classification queries over ontologies.

Abstract

Ontologies are used within Knowledge Representation and Reasoning (KRR) to represent a domain of interest and to assert specific knowledge about the domain. This is done through a class hierarchy and explicit syntactic sentences called axioms, which are made up of concepts, roles and objects. Description Logics (DLs) are a group of knowledge representation languages that can be used to formulate ontologies using similar building blocks. An advantage of using DLs is their ability to support reasoning functionality over the axioms, in order to identify implicit knowledge from the explicitly stated facts. Such reasoning can be performed automatically by software inference engines called reasoners. In a similar way that an ontological concept is defined by declaring facts about the concept, an organism or taxon in taxonomy is defined by specifying all of its unique, defining characteristics. The conceptual process of defining a concept in an ontology, and defining a taxon is similar, thus ontologies can be used to model a taxonomy, and classification can be performed through DL queries. Taxonomy is the scientific classification, description and grouping of certain objects or organisms, and the principles that enforce such classification. One of the goals of taxonomists is the ability to communicate their work, which is normally done through taxonomic keys that are used to identify organisms, and are usually text based. When identifying and grouping objects, certain questions arise such as ‘which objects exist that have various identified unique features?’ and the reverse of the mentioned question, when dealing with the taxonomic process of taxonomic revisions, ‘what features does each (speculated) object possess, and which are the common shared features between them?’ When asking the second question as a query over an ontology, acquiring the needed results proves difficult when using the standard reasoning services. Ways to perform the query through the remodelling of the ontology exist, but are cumbersome and time consuming if dealing with a large ontology. In this dissertation, an alternate way to solve such a query through the use of an existential reasoning algorithm that utilises and extends the standard reasoning services thus avoiding the redundant remodelling, is presented. It is illustrated in a practical way using an ontology and a web ontology based classification application, both which are developed as part of this research study. The ontology and application together function as a computerised taxonomic tool for a specific case study of Afrotropical bees, though they can be applied and used in other domains.