Multi-Level Modeling

Multi-Level Modeling

Multi-Level Modeling (MLM) is a difficult and challenging task. Research activities discuss a lot of methods and concepts to get a handle on the many problems arising in the field. Topics of discussion among others include Power Types(PTs), Partitioning Classes(PCs), categorization, MP, mode reification, shallow, deep and dual deep instantiation, Regularity Attributes(RAs) and their use in deep instantiation (DI) with respect to the attributes durability and mutability. Also a big portion of the discussion is dedicated to the organization of Knowledge Subjects (KS) into layers.

Along with the complexity of concepts which a modeler has to handle comes the aspect that guidelines for MLM are complicated, difficult to understand and apply, or even contradictory or missing. Certainly, there often are alternatives in concrete cases, but then questions come up like „What is the best variant of model?“ or „Which guidelines can I apply to transform a ontology into a better one?“ or „How can I assure reusability and interoperability with other MLMs?“.

Aims: The modeling methods and guidelines should be generally applicable to formal ontology engineering, object-oriented programming (OOA, OOD, UML), development of large-scale database applications and expert systems, and applications of the semantic web (RDF/RDFS/ OWL) [RDF]. Since these disciplines have different vocabularies for concepts, the paper tries to align with standards and terminologies as much as possible. It tries to avoid known deficits in the methods and technologies used. The application of the proposed methods and guidelines will lead to models that are leaner, less redundant, easier to understand and to maintain, more correct and interoperable/compatible and more powerful without, on the other hand, losing semantics or expressiveness compared to methods used so far. In concrete, in this section we will discuss an extended approach to MLM:

• MLM can be significantly improved by merging a Power Type (PT) with its base class, e.g.,(Car, Car_Model) → (Car). This method is coined PTS for PowerType Absorbance.
• The materialization pattern(MP), which is used in combination with PTs, will be resolved in a way that relationships between the pair of powertypes classes and their instances can be saved by PTA.
• Relators will be introduced to model n-ary relationships adequately and to use them for the implementation of LPGs.
• Objects Properties (OP) will be respecified to act as first-class citizens. It will be discussed how this affects their dual facet behaviour (2FB) with respect to punning. The aim is to avoid punning at all.
• In RDF,rdf:type is used at least in two different ways: it can express the instantiation of an individual from a class like(Harry_the_eagle, rdf:type, golden_eagle) and can be used for the instantiation of a metaclass like(golden_eagle, rdf:type, species). It will be discussed, how the different uses ofrdf:type can be modeled in a less mis-interpretable fashion.
• At the end, the organization of KEs in layers should become more transparent. Layer mistakes will be avoided. The modeler should as far as possible be freed from the decision in what layer to place KSs or when to model KEs as classes or instances [BeHu2021].

Motivation: The development of good ontologies as proposed in [GuWe2004a,JaSm2008a,ScSe2012] should be feasible even for subject matter experts who do not have a broad background and experience in mathematics, ontologies and description logic(DL). Throughout the modeling process, modelers should not be forced to always keep issues in mind such as predictability, scalability and performance. The learning curve for acquiring conceptual modeling expertise will be much steeper, as fewer concepts and guidelines need to be learned. The development process will be less error-prone, as feedback on the correctness of the models can be derived from their rich graphical visualizations (GV). The modeler is supported in avoiding layer errors, bad and erroneous designs and errors such as category errors, value constraints and transitive role errors, complex domain/area constraint errors, physical granularity errors, errors regarding unfeasible feasibility and non-referencing information units [ScSe2012] or errors like having cycles in class hierarchies.

Structure of this section: In section MLM Related work, an overview of previous and related work in the field of MLM is given, and various issues arising from the application of different methods and tools of conceptual modeling in the field of structural ontologies are discussed. Also, the concepts of powertypes and regularity attributes are analyzed. In section Species Ontology, using the Species Example Ontology SEO, the application of our MLM methods are described. The discussion in section MLM Related Work relates our MLM approach to others and shows that other MLM requirements can be fulfilled. The final conclusion section MLM Summary summarizes the benefits of our approach.