Although research towards the reduction of the knowledge acquisition bottleneck in ontology engineering is advancing, a central issue remains unsolved: Light-weight processes for collaborative knowledge engineering by a massive user base. In this demo, we present HANNE, a holistic application that implements all necessary prerequisites for Navigational Knowledge Engineering and thus reduces the complexity of creating expressive knowledge by disguising it as navigation. HANNE enables users and domain experts to navigate over knowledge bases by selecting examples. From these examples, formal OWL class expressions are created and re ned by a scalable Iterative Machine Learning approach. When saved by users, these class expressions form an expressive OWL ontology, which can be exploited in numerous ways: as navigation suggestions for users, as a hierarchy for browsing, as input for a team of ontology editors.
Over the past years, structured data has become widely available. Still, the retrieval of dedicated knowledge for given applications or research questions out of these data sources remains a tedious process. A domain expert might have a very precise idea of the concepts she would like to retrieve from a knowledge source. Yet, she faces a number of challenges when trying to retrieve corresponding examples out of a particular data set.
Due to their sheer size, users of these knowledge bases can hardly know which
identi ers are used and are available for the construction of queries. Furthermore,
domain experts might not be able to express their queries in a structured form
at all, but they often have a very precise imagination what kind of results they
would like to retrieve. A historian, for example, searching in DBpedia [
The construction of queries asking for objects of a certain kind contained in an ontology, such as in the previous example, can be understood as a class construction problem: We are searching for a class expression which subsumes exactly those objects adhering to our informal query (e.g. ancient Greek law philosophers in uenced by Plato 1).
In recent years, several methods have been proposed for constructing
ontology classes by means of Machine Learning techniques from positive and negative
examples (see [
In this paper, we present HANNE - a Holistic Application for Navigational kNowledge Engineering. HANNE allows for the extraction of formal de nitions of user-de ned concepts and the corresponding examples out of arbitrary and possibly large RDF data sets. Based on initial examples given by the user, HANNE learns a formal OWL Class Expression of the concept that the user is interested in. This expression is converted into a SPARQL query2 and passed to a triple store database with reasoning capabilities. The results are gathered and presented to the user to choose more examples, to re ne the query, and to improve the formal de nition at will.
Our tool, available online at http://hanne.aksw.org, addresses and circumvents the barriers to the acquisition of knowledge out of data sets: (1) it does not need any deployment and provides a user interface in a familiar surrounding, the browser, (2) the meaning of the identi ers used in the knowledge source is made explicit by the tool, and, nally, (3) the application uses OWL; the results are thus represented in a readable, portable and sustainable way. 2
At the time of writing, the DBpedia ontology class http://dbpedia.org/ontology/ Country contained 2505 instances, including all current countries as well as all historic countries, most of which ceased to exist nowadays.
On April 27th, 2010, there has been a discussion on the DBpedia mailing list3 on how to retrieve (via SPARQL) a list of current countries only, as the coverage of the OWL class was obviously too imprecise (or its de nition was too broad). One suggested solution4 by a DBpedia expert was to manually include a lter for dbo:dissolutionYear 5 within the SPARQL query. 1 technically we mean OWL class expressions such as AncientGreekPhilosopher and in uencedBy value Plato in Manchester OWL Syntax http://www.w3.org/TR/ owl2-manchester-syntax/ 2 http://www.w3.org/TR/rdf-sparql-query/ 3 http://www.mail-archive.com/dbpedia-discussion@lists.sourceforge.net/ msg01652.html 4 http://www.mail-archive.com/dbpedia-discussion@lists.sourceforge.net/ msg01658.html 5 http://prefix.cc/dbo
Although, the request (originally posted by a DBpedia user) was answered, two shortcomings remain: 1. The answer was not recorded or documented in a sustainable way (e.g. incorporated as OWL class within the ontology) 2. The process of nding the answer was very tedious for the user. He had to wait several days and required the help of an ontology expert that was familiar with the existing vocabulary.
In the following, we will explain step by step, how an OWL class (named e.g. Real Country ) can be created without hardly any e ort and previous knowledge with HANNE. On the left side of Figure 1, a full text search over the DBpedia data set can be conducted. This represents the entry point, as initial examples have to be chosen to bootstrap the learning process. In our case, a user could start by searching for \Germany". From the search result, she picks Germany as a positive example and East Germany, West Germany, Nazi Germany as negatives. After she has pressed the learn button (middle, above given examples) a formal OWL de nition (in Manchester OWL Syntax) is presented in the top middle (Learned Concept ) in this case http://www.opengis.net/gml/ Feature and dbp:sovereigntyType some Thing . She now has two options on how to proceed: 1. if she nds the learned concept adequate, she can label (e.g. Real Country ), comment (Countries, which are o cially accepted and still exist ) and save it to export a complete list of instances 2. Alternatively, she can retrieve instances matching the learned OWL class, which are then displayed on the left side Classi ed Instances. These instances can be further evaluated and more positive and negative examples can be chosen to iterate the process. In our case, a total of 261 instances adhere to the class de nition, a quite accurate list (manually checked, including some cases, such as the Azores or the Isles of Man, which are arguable). 3
The application6 realizes a holistic approach to Navigational Knowledge
Engineering, as it combines navigational features with knowledge engineering
capabilities. It is implemented in Java based on the Google Web Toolkit7 and is
made up of highly con gurable and extensible Spring components, so that it can
be customized and tailored for certain data sets. The default implementation of
the component interfaces is held generic and works on arbitrary SPARQL
endpoints with RDFS-Reasoning capabalities8. The full text search (Figure 1 left
side) is based on a con gurable SPARQL template engine. For learning OWL
class expressions, DL-Learner9[
To help users understand the meaning of learned class expression, labels and comments are displayed in a tooltip, when hovering over a named class or property. Advanced users or ontology editors can also manually alter the class expression by selecting suggested classes, which are either more special or more general than the currently learned example. Whenever the learned class expression changes an additional reasoner is queried and shows related concepts from the formerly saved class expressions, which are either sub-, super-, or sibling classes. All saved classes can be browsed and loaded by all users to further re ne searches. If all classes are exported in bulk, they form a class hierarchy, which can be utilized as additional schema for browsing or as input for a team of ontology engineers.
At the time of writing, we con gured the Web demo for DBpedia and a Linguistic data set, but plan to increase the number of available knowledge bases. 6 source code available at http://nlp2rdf.googlecode.com 7 http://code.google.com/webtoolkit 8 our local mirror of DBpedia used in the demo is Virtuoso based http://virtuoso.
openlinksw.com/ 9 http://dllearner.org