Tool development for and empirical experimentation in OWL ontology research require a wide variety of suitable ontologies as input for testing and evaluation purposes and detailed characterisations of real ontologies. Findings of surveys and results of benchmarking activities may be biased, even heavily, towards manually assembled sets of \somehow suitable" ontologies. We are building the Manchester OWL Repository, a resource for creating and sharing ontology datasets, to push the quality frontier of empirical ontology research and provide access to a great variety of well curated ontologies.
Empirical work with ontologies comes in a wide variety of forms, for example
surveys of the modular structure of ontologies[
Pool Interface
BioPortal OWL/XML ORIGINAL
OWL/XML Web Frontend
Web crawl Curation
MOWLCorp
OWL/XML ORIGINAL Manchester OWL Repository
Restful Webservice Oxford Ontology
Library
OWL/XML ORIGINAL ORIGINAL
API no shared understanding of what it means to \use BioPortal". Di erent authors have di erent inclusion and exclusion criteria, for example they only take the ones that are easily parseable after download, or the ones that were accessible at a particular point in time. The Manchester OWL Respository aims to bridge that gap by providing a framework for conveniently retrieving some standard datasets and allowing users to create, and share, their own. 2
The Manchester OWL repository can be divided into four layers (see Figure 1). The rst layer represents the data gathering. Through web crawls, web scrapes, API calls, and user contributions ontologies are collected and stored in their respective collections. The second layer represents the three main data sources of the repository, each providing ontologies in their original and curated (OWL/XML) form. The third layer, the pool, represents a virtual layer in which access to the ontologies is uni ed, providing some means for de-dupli cation because of the possibility of corpora intersection. Lastly, the interface layer provides access to the repository through a REST service and a web-based front end. 3
The main component of the data gathering layer is a web crawl based on crawler4j, a java-based framework for custom web crawling and daily calls to the Google Custom Search API that lls the MOWLCorp, which makes up the bulk of the repository's data. An ongoing BioPortal downloader creates a snapshot of BioPortal once per month using the BioPortal web services, whilst retaining copies of all available versions so far. The third (minor) component of the repository is a web scrape of the Oxford Ontology Library (OOL), a hand curated set of ontologies which features some particularly di cult, and thus interesting to reasoner developers, ontologies. Ontologies are downloaded in their raw form and thrown in the curation pipeline. 4
Ontology candidates from all three sources undergo a mild form of repair
(undeclared entity injection, rewrite of non absolute IRIs) and are exported into
OWL/XML, with their imports closure merged into a single ontology, while
retaining information about the axiom source ontology through respective
annotations. Metrics and les for both the original and the curated versions of
the ontologies are retained and form part of the repository. The data curation
looks slightly di erent for all three data sources, especially with respect to
ltering. Apart from the criterion of OWL API [
There are currently three di erent means to access the repository: 1) A web frontend1 provides access to preconstructed datasets and their descriptions, 2) an experimental data set creator allows users to create custom datasets based on a wide range of metrics and 3) an experimental REST-based web service that allows users to create a dataset using the REST API. Since 2) is based on 3), we now describe the query language that allows users to create their own datasets and access the web service.
The query language allows the user to construct statements that represent lter criteria for ontologies based on some essential metrics such as axiom and entity counts, or pro le membership. It roughly conforms to the following grammar:
q = comp f("&&"j"jj") compg comp = metric (\>=" j \<=" j \=") n metric = \axiom count" j \class count" j ... where \metric" should be a valid metadata element. The query language parser was built with open-source parser generator Yacc and Lex.
The repository web services are built using the PHP framework Laravel. Laravel is an advanced framework which implements the REST protocol, so that users can get access to the services using a REST client, or simply using a web 1 http://mowlrepo.cs.manchester.ac.uk/
method param return service query url /api/ check status /api/checkStatus/ GET download /api/resource GET
POST query JSON array with elds:status, count,size,
message, progress id JSON array with elds: status, progress id le stream browser and web-based tools such as Curl. For now, we have implemented three services: query, checkStatus and download. The query service accepts a query string that complies to the query language and returns an id string. Afterwards, users can use the id string to check the status of their query, and to download the nal dataset using checkStatus and download services.
The usage of the services are listed in the Table 1; note that urls should be appended to mowlrepo.cs.manchester.ac.uk which has been omitted. 6
We have presented the Manchester OWL Repository and a range of prototype interfaces to access pre-constructed datasets and create custom ones. We believe that the repository will help pushing the quality frontier of empirical ontologyrelated research by providing access to shareable, well curated datasets. We are currently working on the REST services, the dataset creator and improved dataset descriptions. In the near future, we are aiming to 1) integrate the repository with Zenodo, a service that allows hosting large datasets that are citable via DOIs, 2) extend our metadata to capture even more ontology properties (in particular consistency and coherence) and 3) improving the curation pipeline by implementing extended yet save xes for OWL DL pro le violations.