=Paper=
{{Paper
|id=None
|storemode=property
|title=HANNE - A Holistic Application for Navigational Knowledge Engineering
|pdfUrl=https://ceur-ws.org/Vol-658/paper522.pdf
|volume=Vol-658
|dblpUrl=https://dblp.org/rec/conf/semweb/HellmannUL10
}}
==HANNE - A Holistic Application for Navigational Knowledge Engineering==
https://ceur-ws.org/Vol-658/paper522.pdf
HANNE - A Holistic Application for
Navigational Knowledge Engineering
Sebastian Hellmann, Jörg Unbehauen, Jens Lehmann
AKSW Research Group, http://aksw.org, Universität Leipzig, Germany
lastname@informatik.uni-leipzig.de
Abstract. Although research towards the reduction of the knowledge
acquisition bottleneck in ontology engineering is advancing, a central is-
sue 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 Nav-
igational 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 refined by a scalable Iterative Machine Learning ap-
proach. 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.
1 Introduction
Over the past years, structured data has become widely available. Still, the re-
trieval 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
identifiers 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 [2] for
ancient Greek law philosophers influenced by Plato can easily name some exam-
ples and if presented a selection of prospective results she will be able to quickly
identify false results. However, she might not be able to efficiently construct a
formal query adhering to the large DBpedia knowledge base a priori.
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 influenced by Plato 1 ).
In recent years, several methods have been proposed for constructing ontol-
ogy classes by means of Machine Learning techniques from positive and negative
examples (see [3] for an overview). Due to their dependency on reasoning meth-
ods, these techniques are tailored for small and medium size knowledge bases
and cannot be directly applied to large knowledge bases. The scalability of the
algorithms is ensured, however, by reasoning only over ”interesting parts” of a
knowledge base for a given task [1]. As a result users of large knowledge bases are
empowered to construct queries by iteratively providing positive and negative
examples to be contained in the prospective result set.
In this paper, we present HANNE - a Holistic Application for Navigational
kNowledge Engineering. HANNE allows for the extraction of formal definitions of
user-defined concepts and the corresponding examples out of arbitrary and pos-
sibly 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 pre-
sented to the user to choose more examples, to refine the query, and to improve
the formal definition at will.
Our tool, available online at http://hanne.aksw.org, addresses and circum-
vents 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 identifiers used in the knowledge source is
made explicit by the tool, and, finally, (3) the application uses OWL; the results
are thus represented in a readable, portable and sustainable way.
2 Example Usage
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 definition was too
broad). One suggested solution4 by a DBpedia expert was to manually include
a filter for dbo:dissolutionYear 5 within the SPARQL query.
1
technically we mean OWL class expressions such as AncientGreekPhilosopher and
influencedBy 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
Fig. 1. Screenshot of the left and middle part of http://hanne.aksw.org: Real
countries in DBpedia. (The right part containing a list of stored concepts and
additional features is omitted for a larger image and readability)
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 finding 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 effort 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 neg-
atives. After she has pressed the learn button (middle, above given examples)
a formal OWL definition (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 finds the learned concept adequate, she can label (e.g. Real Country ),
comment (Countries, which are officially 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 Clas-
sified 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 definition, a quite accurate list (manually checked,
including some cases, such as the Azores or the Isles of Man, which are arguable).
3 Overview of the Application
The application6 realizes a holistic approach to Navigational Knowledge Engi-
neering, as it combines navigational features with knowledge engineering capa-
bilities. It is implemented in Java based on the Google Web Toolkit7 and is
made up of highly configurable 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 end-
points with RDFS-Reasoning capabalities8 . The full text search (Figure 1 left
side) is based on a configurable SPARQL template engine. For learning OWL
class expressions, DL-Learner9 [3] is used. [1] describes the underlying technique
of machine learning on large knowledge bases and contains performance mea-
surements (especially on DBpedia) with acceptable speed for a web scenario.
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 refine
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 configured the Web demo for DBpedia and a
Linguistic data set, but plan to increase the number of available knowledge
bases.
References
1. Sebastian Hellmann, Jens Lehmann, and Sören Auer. Learning of OWL class de-
scriptions on very large knowledge bases. IJSWIS, 5(2):25–48, 2009.
2. Jens Lehmann, Chris Bizer, Georgi Kobilarov, Sren Auer, Christian Becker, Richard
Cyganiak, and Sebastian Hellmann. DBpedia - a crystallization point for the web
of data. Journal of Web Semantics, 7(3):154–165, 2009.
3. Jens Lehmann and Pascal Hitzler. Concept learning in description logics using
refinement operators. Machine Learning journal, 78(1-2):203–250, 2010.
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