Multilevel modeling aims at improving the expressiveness and conciseness of conceptual modeling languages by allowing to express domain knowledge at higher abstractions levels. In this demonstration, we go thru two variants of multilevel extensions for the ConceptBase system, which had originally been used more for the design of domainspeci c conceptual modeling languages. The demonstration highlights the partial evaluation feature of the deductive rule engine of ConceptBase. It also shows how multilevel modeling is essentially about a better understanding how instantiation, specialization, and attribution relate to each other in conceptual modeling.
This demonstration is showcasing the multilevel modeling capabilities of
ConceptBase [
ConceptBase was originally developed as a repository for software engineering artefacts. It is technically a deductive database with its own Datalog-neg engine to evaluate recursive deductive rules. The data model is a variant of the Telos language, which itself is based on the three principles instantiation, specialization, and attribution (subsuming relations between objects). Telos has a potentially unlimited instantiation hierarchy (individuals, classes, metaclasses, metametaclasses, etc.) terminated by a so-called omega-level class "Proposition" (standing for all objects), which has any explicit information as instance. This includes explicit attributes, instantiations, and specializations, which all are objects in Telos. ConceptBase also features a customizable graphical user interface CBGraph. It displays views on the database (i.e. models) by representing nodes and links by graphical shapes depending on (derived) properties of the objects. For example, a deductive rule can specify to display instances of the class "RelationshipType" with a diamond shape. The textual editor CBIva allows to de ne new objects in a frame syntax and query the ConceptBase database.
Traditionally, ConceptBase was mainly used for metamodeling, in particular to de ne families of interrelated domain-speci c modeling languages. Deductive rules and integrity constraints are used to specify well-formed models and the semantics of modeling constraints. For example, the semantics of cardinality constraints can be expressed by integrity constraints (a special form of deductive rules). The multilevel modeling case is a particular challenge for the deductive rule engine of ConceptBase, since it has to cope with objects at more than two abstraction levels. Further, multilevel modeling has an intricate relation to the core principles of conceptual modeling, i.e. instantiation, specialization and attribution. These principles are prede ned in the ConceptBase implementation of Telos.
The rst part of the demonstration discusses our rst attempt to implement constructs for multilevel modeling with ConceptBase. Dual deep instantiation (DDI) is a so-called potency-based approach, where attributes and (binary) relations have a source and a target potency. When both numbers are set to "1", then the attribute/relation behaves like a in classical conceptual modeling: the attribute/class is de ned at the class level and is instantiated at one level below the current class level. When the numbers are greater than "1", then the attribute/relation is instantiated at lower levels depending on the number. For example a metaclass "ProductType" could have an attribute "serialnumber" with source potency of 2 and target potency of "1". Assume that "Car" is an instance of "ProductType". Then, an instance "mycar123" of "Car" can have a serial number, without any need to de ne it for the class "Car". It is de ned for all products. The demonstration consists of the following steps: 1. De ne the metaclass "ProductType" and its "serialnumber" attribute with the desired potencies. 2. De ne a class "Car" with a regular attribute "numberofdoors". 3. De ne an instance like "mycar123" that instantiates both attributes. 4. De ne another class "MobilePhone" with attribute "protocol". Instantiate "MobilePhone" and re-use the attribute "serialnumber".
We shall also show the formulas de ning DDI and how they are compiled to two-level formulas by the partial evaluator of ConceptBase. Dual deep instantiation provides a number of additional capabilities to constrain the use of attributes and relations de ned at higher instantiation levels. In particular, attributes/relations with potencies can be restricted at lower instantiation levels. The second part of the demonstration uses the same example but is based on the powertype construct. Rather than using level numbers, it encodes the multilevel structure of a model by in instantiation hierarchy of powertypes. The attributes/relations are then de ned at the suitable level of this hierarchy. Multiple hierarchies can co-exist and be linked via relations. The demonstration consists of the following steps: 1. De ne the metaclass "ProductType" as powertype of "Product". 2. De ne a class "Car" as instance of "ProductType". De ne the attribute "serialnumber" for "Product" and "numberofdoors" for "Car". 3. De ne an instance like "mycar123" that instantiates both attributes.
In this simple scenario, the MLT-Telos variant is more concise than DDI but lacks the expressiveness of the dual levels of attributes/relations in DDI. We shall demonstrate also some larger examples from the web site for this demo.
We demonstrate the role of partially evaluating rules that range over more than 2 levels to sets of rules that range over two levels, see gure 3. The multilevel rule considered here is forall t1,t2/Proposition e/Proposition
(t1 specializes_1 t2) and (e in t1) ==> (e in t2)
This rule has variables e, t1 and t2, where e is an instance of t2, if t1 specializes t2 and e is and instance of t1. The "specializes" relation is de ned as the level of the class "Proposition". So we have three levels of objects here: Proposition is at the highest level, t1 and t2 are at the middle level, and the variable e ranges over objects at the lowest level. Now, when we have a (derived) fact like (Car specializes 1 Product), then we can substitute t1 with "Car" and t2 with "Product" and we can generate a partially evaluated 2-level formula: forall e/Car (e in Product)
ConceptBase automatically generates the 2-level formulas, which are much more e cient to evaluate due to the lower number of variables and thus reduced number of join operations for the Datalog engine. 6
This paper provided the background for the two multi-level modeling variants DDI and MLT-Telos, both being implemented by the ConceptBase system via its metamodeling and deductive capabilities. The two demonstration examples highlight the di erences between the two approaches. In short, MLT-Telos is better integrated with the Telos axioms by reusing its notions for instantiation and specialization. DDI has an elaborate system of rules and constraints expressing how attributes and relations can be specialized and instantiated. The main purpose of the demonstration is to explain how a collection of metaclass de nitions plus a set of deductive rules and integrity constraints can capture most of the desired semantics of multilevel modeling paradigms. The web page http://conceptbase.sourceforge.net/er21demo contains further instructions to run the demonstration on your own computer, as well as links to videos of the demonstration.