Ontology evolution is an essential research area for the widespread use of ontologies in industrial and academic applications. In this paper, we present Change Management Patterns CMP defining three kinds of patterns: Change Patterns, Inconsistency Patterns and Alternative Patterns. CMP patterns are proposed to guide ontology evolution process by driving and controlling change application while maintaining consistency of the evolved ontology.
Ontology development is a dynamic process starting with an initial rough ontology,
which is later revised, refined and filled in with the details [
Ontology evolution is a complex problem. In our work1, we focus on issues related to ontology change management in a local context and particularly on consistency maintenance. To guide ontology evolution, we have defined Change Management Patterns CMP. The patterns model the three dimensions: Change, Inconsistency and Resolution Alternative. Based on the modeled patterns and the conceptual links between them, we propose an automated process driving change application while maintaining consistency of the evolved ontology.
The paper is organized as follows: in section 2, we detail the application context of Change Management Patterns (CMP) which are described and illustrated in section 3. Application of CMP to guide change resolution is presented in section 4. Before concluding and discussing further developments of this work, we report on related work in section 5. 2 CMP as a Meta-layer on Top of an Ontology Evolution Process CMP patterns are proposed as “meta-layer” of an ontology evolution approach – ONTO-EVOAL – guiding the change management process at three key phases: change specification, change analysis and change resolution (Fig. 1). Three categories of patterns are modeled: Change Patterns classifying types of changes, Inconsistency Patterns classifying types of logical inconsistencies and Alternative Patterns classifying types of inconsistency resolution alternatives.
The starting point of ONTO-EVOAL process is the specification of the required change by formally describing its semantics (intermediate changes composing it –if any– and their order, involved entities, change values, etc.). This is guided by the instantiation of the corresponding Change Pattern, resulting in an explicit change signature. The following phase aims to analyze change impact by explaining and localizing caused inconsistencies. Detected inconsistencies are classified according to the corresponding Inconsistency Patterns. In the change resolution phase, the instantiation is not made from process level to pattern level, but rather in the other way i.e. by generating alternative instances from the conceptual links between inconsistency patterns –instantiated in the previous phase– and Alternative Patterns resolving them (Section 3). Resolution alternatives represent additional and/or substitutive changes to implement to maintain ontology consistency.
Pattern Level Process Level Initial Ontology Required Change Historic Level
Change Management Patterns (CMP) Change Patterns
Inconsistency Patterns
Alternative Patterns
Change Specification
Change Analysis
Change Resolution
Change Application Change Signature
Detected Inconsistencies Resolutions
Final Changes Evaluation Module Quality Model NCCoPmplexPNi1tPéyCCoNhéeRsiCoP2n SNtTrCautRcxtonrePmH3iy-IesAAREPPébvD4MsstuArolatycaiidoo’nénNvaMRlPue5MstiooldnuPlarRreiEctéyPC6CRoaEmpCpUelsétanugPde7seCAComPR8pAréhensNioTnC
Impact of Resolutions
Evolved Ontology Evolution
Results
Evolution Log
In addition, the process includes a Quality Evaluation Module assessing the impact of proposed resolutions on ontology quality. The goal of the evaluation module is to help selecting the resolution that preserves ontology quality. If quality level is maintained, required change and its derived changes are directly applied and the ontology evolves. In the case that all the proposed resolutions have negative effect on ontology quality, the results of the different phases are presented to ontology engineer –as a complement to his expertise– so that he can decide about the changes.
All the results of the process are saved in the evolution log to keep ontology
evolution historic. Besides controlling ontology evolution, revoking or justifying
changes, the evolution log facilitates learning new change management patterns. For a
detailed description of ONTO-EVOAL evolution process, the reader can refer to [
In this section, we want to highlight the role of CMP in guiding the evolution process and more precisely change specification, analysis and resolution phases. CMP are proposed as a solution looking for invariances in change management that repeatedly appear when evolving ontologies. Three categories of patterns are distinguished: Change Patterns, Inconsistency Patterns and Alternative Patterns. The goal of CMP modeling is to offer different levels of abstraction, to establish conceptual links between these three categories of patterns (Fig. 2) determining the inconsistencies that could be potentially caused by a type of change and the alternatives that possibly resolve a kind of inconsistency and thus, to guide an automated change management process.
Change Management Pattern
CMP
CMP classify categories of changes based on OWL meta-model, categories of logical inconsistencies considering OWL DL constraints and categories of inconsistency resolution alternatives. To describe the three categories of patterns and explain theirs relations, we illustrate step-by-step the instantiation of the patterns based on a simple example given as a thread.
Let’s consider the OWL ontology O defined by the following axioms: {Animal ! Fauna-Flora, Plant ! Fauna-Flora, Carnivorous-Plant ! Plant, Plant ! !Animal}
And let’s consider a required change Ch1 defining Carnivorous-Plant class as a sub-class of class Animal.
CMP are modeled as an OWL DL ontology describing CMP catalogue. In addition,
a simplified representation is used. It extends the representation of Ontology Design
Patterns ODP [
CMP are described by a general template containing the following slots: ! General Properties: General Information about the pattern, and Use Case of the pattern; ! Specific Properties related to CMP abstraction level including general properties of Pattern Description (some independent from pattern category and others extended to the different kind of CMP patterns), Graphical Representation of the pattern represented in UML format, and Implementation of the pattern in OWL DL language (not presented in this paper for page limit reason); ! Properties about Relationships of the pattern with other patterns including Relations to Other CMP.
These slots are presented in the following tables (Tab. 1) (Tab. 2) (Tab. 3) (Tab. 4).
3.1 Change Patterns
Change patterns aim to categorize changes, formally define their signification, their
scope and their potential implications. Two categories of OWL changes are
distinguished in the literature [
Basic change patterns describe all OWL basic changes derived from OWL metamodel. They model simple and indivisible OWL changes. To extend the set of defined OWL basic operations with composite and specific change operations, we define complex change patterns. They correspond to composite OWL changes grouping logical sequences of basic and composite changes (e.g. enlarging the range of an object property to its super-class or merging two classes).
Considering the example given above, the change Ch1 instantiate the basic change pattern: Add a sub-class. The description and the instantiation of the pattern are given by the following table (Tab. 1). CMP Type* Problem Examples Intent Consequences Scenarios
Use Case Define a sub-class relation between two classes.
Suppose that we need to express that the class CarnivorousPlant is a sub-class of the class Animal.
CMP Abstraction Level
Pattern Description The pattern models an indivisible change that defines a subclass relation between two classes and notifies if there is any constraint on this subsumption so that it does not affect the logical consistency of the ontology.
The pattern defines a subsumption between two classes and notifies that the super-class and the sub-class should not be disjointed.
Define the class Carnivorous-Plant as a sub-class of the class Animal.
Notify that the constraint: {class Carnivorous-Plant and class Animal should not be disjointed} has to be verified.
Change Abstraction Level
Pattern Description Basic Change Pattern.
Arguments* ID of the sub-class (sub_classID).
Carnivorous-Plant.
Referred Entities* ID of the sub-class (sub_classID).
ID of the super-class (super_classID).
Examples Carnivorous-Plant.
Animal.
Constraints !(sub_classID disjointWith Super_classID). !(Carnivorous-Plant disjointWith Animal).
Graphical Representation
Super-Class Sub-Class {Super-Class Not Disjoint with Sub-class} Inconsistency Patterns Carnivorous-Plant
{Animal Not Disjoint with Carnivorous-Plant}
Relationships
Relations to other CMP
Inconsistency disjointness related to subsumption. 3.2 Inconsistency Patterns After applying temporary the required change, inconsistencies detected using an ontology reasoner, are classified according to inconsistency patterns. Inconsistency patterns model a sub-set of OWL DL logical inconsistencies: disjointness inconsistencies related to subsumption and instantiation; inconsistencies related to equivalence and complement, inconsistencies related to equivalence and disjointness, inconsistencies related to value restrictions and inconsistencies related to cardinality restrictions.
Reconsidering the example given as a thread, the pattern corresponding to the disjointness inconsistency detected and its instantiation are described by the following table (Tab. 2). ID SuperClass1 ID SuperClass2 ID SubClass ID SuperClass1 ID SuperClass2 ID SubClass ID SuperClass2 ID SubClass ID SuperClass2 ID SubClass Involved Axioms Examples Responsible Axioms Examples Diagram Diagram Examples
Explain disjointness inconsistency caused by a subsumption relation between the class Carnivorous-Plant and the class Animal by tracking the classes concerned by this inconsistency and specifying the axioms causing it.
Inconsistency Abstraction Level
Pattern Description
Arguments*
Implicated Entities* ID of a first super-class (super_class1ID).
ID of a second super-class (super_class2ID).
ID of the sub-class (sub_classID).
Examples
Involved Entities* ID of the involved super-class (super_class2ID). ID of the sub-class (sub_classID).
Examples
Axioms* (super_class1ID disjointWith super_class2ID), (sub_classID ! super_class1ID). (Plant ! !Animal), (Carnivorous-Plant ! Plant). (sub_classID ! super_class2ID). (Carnivorous-Plant ! Animal)
Graphical Representation super_class2ID Change Patterns Alternative Patterns
Relationships
Relations to other CMP Add a sub-class, … - Define Hybrid Class for Resolving Disjointness Subsumption, - Enlarge Class Definition for Resolving Disjointness
Subsumption. 3.3
Alternative Patterns Change resolution is based on the conceptual N-ary relation-class Resolution defined between change, inconsistency and alternative patterns (Fig. 2). For each detected inconsistency, based on the corresponding inconsistency pattern instance and the instantiated change pattern specifying the required change, potential alternative patterns are generated and instantiated. An alternative pattern represents an additional change (applied jointly to the required change) or a substitutive change to apply (replacing the required change) so that a logical inconsistency can be resolved. It is described as a change (basic or complex) and it inherits and extends change pattern properties (Fig. 2).
Several resolution alternatives can be proposed for an inconsistency. To resolve the inconsistency described above in the example, two alternatives can be proposed: the first one (Tab. 3) is a substitutive resolution extending a complex change. The second one is a substitutive resolution extending a basic change (Tab.4).
Alternative Pattern Example
General Properties
General Information Define Hybrid Class for Resolving Disjointness_Subsmption.
AltP_DefHybClsResolDisjSubs Alternative Pattern.
Use Case Resolve disjointness –related to a subsumption– by defining a hybrid class.
Suppose that we need to resolve a disjointness inconsistency caused by subsuming the class Animal by the class Carnivorous-Plant.
CMP Abstraction Level
Pattern Description The pattern models a resolution alternative resolving disjointness inconsistency –related to a subsumption– by creating a hybrid class.
The pattern resolves a disjointness inconsistency –related to a subsumption–by defining a hybrid class based on the definition of disjoint classes implicated in the inconsistency, and redistributing correctly sub-class relations between classes implicated in the inconsistency, the hybrid class, and the most Process Examples Change Pattern Type* Object Type* Involved Entity Type* specific common super-class of the disjoint classes implicated.
Define a hybrid class Animal_Plant based on the definition of the two disjoint classes involved in the inconsistency: Animal and Plant.
Then, create a sub-class relation between the hybrid class created and a the most specific common super-class of the classes Animal and Plant.
And finally, substitute the sub-class relation between the classes Animal and Carnivorous-Plant by a subsumption between the classes Carnivorous-Plant and Animal_Plant.
Alternative Abstraction Level
Pattern Description 1) The pattern defines a hybrid class as a union of the definitions of the disjoint classes implicated in the inconsistency to be resolved; 2) The pattern defines a subsumption between the most specific common super-class of the disjoint classes implicated in the inconsistency and the hybrid class created; 3) The pattern defines a subsumption between the hybrid class and the sub-class involved in the inconsistency. 1) The pattern defines a class Animal_Plant as a union of the definitions of the disjoint classes Animal and Plant; 2) The pattern defines a subsumption between the most specific common super-class of the disjoint classes Fauna-Flora and the hybrid class created Animal_Plant; 3) The pattern defines a subsumption between the defined hybrid class Animal_Plant and the sub-class Carnivorous-Plant involved in the inconsistency.
Change Abstraction Level
Pattern Description Complex change pattern.
Arguments* ID of the hybrid class (HybridClassID).
Animal_Plant.
Referred Entities* ID of the sub-class (sub_classID).
ID of a first disjoint class (Disjoint_Class1ID). ID of a second disjoint class (Disjoint_Class2ID).
Examples Carnivorous-Plant.
Animal.
Plant.
Intermediate Entities* ID of the most specific common super-class of the disjoint classes implicated (Common_super_classID).
Fauna-Flora.
Complex Change Abstraction Level
Pattern Description
Collection(Disjoint_Class1ID, Disjoint_Class2ID), Operator(Union)) ; 2) BChP_AddSubClass (HybridClassID, Common_super_classID) 3) BChP_AddSubClass (sub-ClassID, HybridClassID) 1) BChP_AddClass (Animal_Plant, Collection(Animal, Plant), Operator(Union)) ; 2) BChP_AddSubClass (Animal_Plant, Fauna-Flora) 3) BChP_AddSubClass (Carnivorous-Plant, Animal_Plant) Graphical Representation
Disjoint_Class 2
HybridClass Common_Super_Class Examples Diagram Diagram Examples Change Patterns Inconsistency Patterns Consequences Scenarios Process Examples
Pattern Description The pattern models a resolution alternative resolving disjointness inconsistency –related to a subsumption– by enlarging the definition of a class.
The pattern resolves a disjointness inconsistency –related to a subsumption–by enlarging the definition of the sub-class involved in the inconsistency, based on the definition of disjoint classes implicated in the inconsistency.
Enlarge the definition of the class Carnivorous-Plant based on the definition of the classes Animal and Plant.
Alternative Abstraction Level
Pattern Description 1) The pattern enlarges the definition of the sub-class involved in the inconsistency by adding –in its description– a union of the definitions of disjoint classes implicated in the inconsistency. 1) The pattern enlarges the definition of the sub-class Carnivorous-Plant by adding –in its description– a union of the definitions of the disjoint classes implicated in the inconsistency: the classes Animal and Plant.
Change Abstraction Level
Pattern Description Basic change.
Arguments* ID of the class to enlarge (classID).
Carnivorous_Plant.
Referred Entities* ID of the class to enlarge (classID).
ID(s) of the classes of the collection (Disjoint_Class1ID, Disjoint_Class2ID).
Examples Carnivorous-Plant.
Animal, Plant.
Operator*
Graphical Representation
Common_Super_Class
The primary objective of ontology pattern modeling is to provide shared and reusable
guidelines [
In the change specification phase (Fig. 1), an appropriate change pattern is selected
and instantiated to specify explicitly a required change. The temporary application of
the change is then, analyzed to detect caused inconsistencies. This phase is performed
by employing Pellet Reasoner [
Inconsistency detection and localization prepare inconsistency pattern selection and instantiation. Localized inconsistencies are matched to inconsistency patterns to be classified. The matching process consists in identifying correspondences by considering inconsistency pattern type and structure, and also the semantics of pattern arguments and axioms (Tab. 2), which define the pattern interface that has to be instantiated. Guided by the instantiated change pattern and its constraints (Tab. 1), the matching process targets firstly, inconsistency patterns that could be potentially caused (Fig. 2). Two types of indicator are considered: structural information (matching of sub-hierarchies), and axiomatic information related to localized inconsistencies (localization context, entities declared as inconsistent, axioms related to these entities, etc.).
Once localized inconsistencies are classified, potential alternative patterns are identified and generated based on N-ary relation classes – Resolution – defining semantic resolution relations between instantiated, change and inconsistency patterns. Then, the instantiation of the generated alternative patterns is adapted to the subontology concerned by its application.
Alternative instances proposed for the different inconsistencies are combined to derive global potential resolutions for the required change. Each resolution defines a set of derived changes but should not cause other inconsistencies. Therefore, all derived global resolutions are verified using Pellet reasoner, and only consistent ones are accepted for the evaluation step (Fig.1).
Discussion with related work is tackled through three parts: pattern modeling, ontology evolution approaches, and inconsistency diagnosis and repair.
Pattern modeling was adopted in web ontology design to propose guidelines and
provide reusable ontological component catalogue2. CMP patterns are close to
Ontology Design Patterns3 ODP, particularly Logical Ontology Patterns LOP and
Content Ontology Patterns COP [
Concerning OWL ontology evolution approaches, in [
Concerning inconsistency diagnosis and repair, several debugging services and
strategies are proposed in the literature [
In this paper, we have presented CMP patterns proposed as “meta-layer” of an ontology evolution approach – ONTO-EVOAL – to guide and control the change management process at three key phases: change specification, change analysis and change resolution.
Currently, we are developing a learning module enriching and enhancing CMP by considering evolution log information, new change compositions not yet supported by
3 http://ontologydesignpatterns.org/ Change Patterns, detected inconsistencies not yet classified by Inconsistency Patterns, new potential resolution alternatives not yet modeled by Alternative Patterns, and also new possible relation instantiations between CMP.