- OntoCheck -

Verifying Ontology Naming Conventions & Metadata-completeness in Protégé 4

Motivation

With the advent of the semantic web and RDF-based knowledge representation techniques of-the-shelf ontology editors like Protégé 4 gain widespread use. Although it's functionalities are sufficient for the daily ontology editing tasks, some clean-up checks on the generated ontology - e.g. to be stored and carried out before each ontology release - could complement P4 in a useful way.
We here introduce a new Protege Tab plugin that checks certain properties of an active ontology (OntoCheck) and allows for improvements in the areas of

a) Metadata completeness, e.g. via cardinality checks on mandatory and obligatory annotation properties
b) Naming Conventions, e.g. via lexical analysis and labeling enforcement for representational units (RU) names and IDs.

This work is based on pervious efforts in lexical harmonization, i.e. the OBO Foundry Naming Convention proposal.
For a review of ontology naming conventions, please look at the Survey-based naming conventions for use in OBO Foundry ontology development paper.

Download & Installation

Find the OntoCheck plugin for Protege 4.x for download at:

OntoCheck P4 plugin

Place this jar file in your Protege 4.x plugin folder, then start Protege and open an owl ontology. Make the new OntoCheck Tab visible by checking it in the Window/Tabs menu.
In case you get a distorted look of the plugin tab window, try to configure the tabs borders, so that there is just enough space to show the three panels in the OntoCheck view. E.g. try move the border of the class hierarchy browser to the left. If this does not help, try to select "Window/Reset selected Tab to default state" and re-try. In case this does not work either, or you get a 'Can't load plugin error', try to deactivate the tab and re-activate it by windows/Views/Class Views and drop it into the Field right of the hierarchy pane.

You can store your generated Checks as well as its result lists onto your harddrive. Previously the code of the stored xml was to be found in the OntoCheckSaves folder within your P4 installation directory, e.g. in:
C:\Program Files\Protege_4.1\OntoCheckSaves. For these versions you need to make sure you log into your operating system with write-rights.

Documentation

The Plugin is self-explanatory and provides hints when the mouse pointer is placed over an Item in question. Find an overview of its features and example applications in the ppt OBML 2011 talk , or in the OBML 2011 paper ( page 61).

Here you can also look at a test result-table listing checks carried out on six ontologies together with some outcome quantification.

Screenshots

The Check Tab:

The Check Tab of the OntoCheck Protege plugin

Save and re-load of formulated Checks:

Save-Load Function

The Compare Tab:

Check panel

The Statistics Tab:

The Statistics tab of the OntoCheck Protege plugin

Current Limitations & Desired Feature List

At the moment the user has to amend the labels manually, but RUs violating tests could be corrected automatically (OntoCure) in the future, i.e. where possible, do simple syntactic corrections automatically, e.g. correct case and separator conventions for all found violations.

Check Tab
A future version should:

Allow for the exploitation of user specified lists with words representing ‘affixes of interest’ which are to be detected and alerted on when occurring in labels. This would allow top automatize checks e.g. on stop-words, negation indicators, cardinality indicators, taboo words from the metalevel and many more.
Check and alert automatically, when an imported ontology differs in naming conventions from the active ontology.
Check and alert on violations of class-subclass name patterns, i.e. situations where the head noun of a subclass is not adequately related via a taxonomic correspondence to the head noun of its superclass. It was observed that, for multi-token labels, such violations are nearly always caused either by a modeling error, such as confusion of taxonomy with partonomy, or by a bad naming practice, e.g. ‘parsimonious’ omission of the true head noun; this is a consequence of the set-theoretic nature of OWL ontologies.
Check that qualifier terms (differentia) appear before the part being qualified (genus). E.g. ‘NMR_instrument’ in place of ‘instrument_for_NMR’.
Allowed checks could be expanded on relations, e.g. for object properties, check that mutually inverse relations comply name-wise with each other, e.g. that a certain relation prefix implies a corresponding circumfix in its inverse form, as for ‘has_X’ the inverse should be ‘is_X_of’: Using (genetic_information) ‘has_bearer’ (gene) and (gene) ‘is_bearer_of’ (GeneticInformation) would avoid the hard-to-find relation name 'inheres_in' as inverse of 'is_bearer_of'.
rresponding circumfix in its inverse form, as for ‘has_X’ the inverse should be ‘is_X_of’: Using (genetic_information) ‘has_bearer’ (gene) and (gene) ‘is_bearer_of’ (GeneticInformation) would avoid the hard-to-find relation name 'inheres_in' as inverse of 'is_bearer_of'.
Check for naming clashes in equal (synonymous) fields for different classes, e.g. if there is a class with equal labels represented with different IDs.
Check if an imported ontology differs in naming conventions from the active ontology.
Check for violation of class-subclass naming pattern, i.e., situations when the head noun of a sub-class is not adequately related via a taxonomic correspondence to the head noun of its superclass. It was observed that, for multi-token labels, such violations are nearly always caused either by a modeling error, such as confusion of taxonomy with partonomy, or by a bad naming practice, e.g. ‘parsi-monious’ omission of the true head noun; this is actually a natural consequence of the set-theoretic nature of OWL ontologies.

Compare Tab

In the result list both variant forms will be displayed and the found differences could be highlighted.

Count Tab

Detect abundant pre-, in-, suff- and postfixes and list them according to frequency of occurrence. If a postfix occurs often in siblings, a recommendation could be issued to use this postfix in those labels throughout, or as superclass label.
Detect logical operators like AND, OR, NOT in names, e.g. BioTop has CarbohydrateMole-culeOrResidue and OligoOrPolymer. These could be potentially correlated with actual logical definitions and disjoints.
Semantic analysis could probably guide in expressiveness selection, e.g. words indicating cardinality requirements, such as minimal, maximal, exact hint for certain OWL 2 profiles.
Given a reasoner is activated, as part of an expressivity analysis, besides counting the hub-node-to-isolate class ratio, entailment densities and OWL flavor element usage should also be counted. Such metrics would enable the judgment whether a certain semantics or OWL flavor was chosen because it is en vogue or because it is actually needed.

Contact & Questions

OntoCheck is provided free of cost, is published under the terms of the GNU General Public License and is, of course, provided without any warranty.
If there are still features missing, please contact us for new ideas not mentioned above:

Schober at imbi dot universität-freiburg dot de

Acknowledgements

This work was initiated and supervised by Daniel Schober, Implemented and improved by Ilinca Tudose and hosted by Stefan Schulz and Martin Boeker. It was partly supported by the Deutsche Forschungsgemeinschaft (DFG) grant JA 1904/2-1, SCHU 2515/1-1 GoodOD (Good Ontology Design). Thanks to Timothy Redmond for helping with the Protege API.