I3OM – An Iterative, Incremental and Interactive Approach
for Ontology Navigation based on Ontology Modularization
Ricardo Brandão, Paulo Maio and Nuno Silva
Knowledge Engineering and Decision Support Research Center,
School of Engineering, Polytechnic of Porto, Porto, Portugal
Keywords: Ontology-based Navigation, Ontology Modularization.
Abstract: Although ontologies are used to describe a specific domain of interest, they can grow in size exponentially,
compromising their usage. Furthermore, current ontology engineering tools do not effectively support the
data/information visualization and navigation described through large ontologies. To address these issues,
we claim that the experience and results of navigating/browsing ontology-described data can profit from the
modularization of the ontologies underlying the repositories. For that, we propose the I3OM process that
facilitates ontology-oriented navigation and contextualized information retrieval by combining different
ontology modularization techniques into an iterative, incremental and interactive process.
1 INTRODUCTION
Ontologies are seen as an appropriate formalism to
capture and represent the structure and semantics of
data/information in the Web and, therefore, serve as
the backbone of the Semantic Web (Berners-Lee et
al. 2001). However, despite ontologies describe a
specific domain of interest, their size and complexity
tends to increase too (Del Vescovo et al., 2011).
Thus, ontology understandability decreases as its
complexity increases which consequently leads to an
increase in human effort in apprehend and reuse
them (Stuckenschmidt and Schlicht, 2009).
Ontology-supported navigation is a recent
research field that aims to assist the user in
comprehending, searching and retrieving
information from repositories described through
ontologies (Franconi et al., 2010; Motta et al., 2011).
However, current tools do not effectively support the
navigation through ontologies (Dzbor et al., 2006),
especially those inexperienced and non-experts
users.
While ontology modularity (Parent and
Spaccapietra, 2009) partially tackles these issues, the
existing algorithms do not consider the user in the
loop, and thus are not able to fully respond to the
user requirements.
This paper advocates the need to combine the
user expertise and automatic ontology
modularization algorithms in the ontology-supported
navigation process. For that a novel iterative,
incremental and interactive ontology modularization
(I3OM) algorithm is proposed.
Next section details the context and requirements
of the I3OM. Section 3 introduces the benefits of
ontology modularization and the core definitions
applied during the remaining of the paper. In section
4, the proposed I3OM process is described, further
complemented with a walk-through example in
section 5. In section 6, our proposal is compared to
other works. Finally, section 7 summarizes the
contributions and point out next research steps.
2 CONTEXT
The World Search (WS 2009) project aims to
provide an application for a specific domain (e.g.
health care, public administration) that supports
domain experts during their quest for information
resources. These resources are available in multiple
and heterogeneous repositories. A resource is either
(i) a text document, (ii) an user annotation of a (part
of a) document or (iii) a set of facts in a knowledge
base.
During the analysis of requirements, the
development team observed that the users were
interested neither in text-based searches only, nor in
formal queries to the repository. Instead, users are
interested in an elaborated combination of both. I.e.
265
Brandão R., Maio P. and Silva N..
I3OM – An Iterative, Incremental and Interactive Approach for Ontology Navigation based on Ontology Modularization.
DOI: 10.5220/0004145602650270
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2012), pages 265-270
ISBN: 978-989-8565-30-3
Copyright
c
2012 SCITEPRESS (Science and Technology Publications, Lda.)
users want to have the chance to make a query that
includes free-text and semantic specification of
content. This combination is formally captured by
the next function:
′,_′,_,
where:
is a user entered free-text;
is a set of resources (documents,
annotations or facts) in which the user is
interested for. It serves as example of the
resources to retrieve;
_ is a (partial) formal specification of
the required content based on a model,
typically in the form of a set of taxonomy
entities or ontologies entities. It serves as
formal constraints to the query, i.e. only those
resources semantically defined/annotated with
those entities should be retrieved;
′ is the set of resources retrieved to the user
for visualization. This includes both the
text-based retrieved resources (documents) and
ontology-based retrieved resources
(annotations and facts);
_′ is the set of relevant semantic
entities that (i) belong to the formal model
describing the resources, (ii) is representative
of the semantics of the output resources (′)
and (iii) enables the user to further refine the
query.
The core of the problem lays on:
the combination of the different input’ types to
the query;
the required iterative approach that implies not
only the retrieved resource (′) but also a set
of semantic entities (_′) that will
support the refinement of the query.
3 MODULARIZATION
Modularization refers to a situation where a thing
(e.g. an ontology) exists as a whole but can also be
seen as a set of parts (the modules) (Parent and
Spaccapietra, 2009). In the knowledge management
(ontology engineering) scenario, by splitting an
ontology into smaller parts, one is allowing the
selective use of knowledge which (i) facilitates
ontology reusability and share-ability
(Stuckenschmidt and Schlicht, 2009), (ii) reduces
the human effort in understanding such ontology
(Parent and Spaccapietra, 2009), (iii) empowering
the ontology manipulation, maintenance and
evolution tasks (Parent and Spaccapietra, 2009), (iv)
improves the usage of reasoners (e.g. by Distributed
Reasoning, by incremental reasoning) (Del Vescovo
et al., 2011). Yet, another advantage of ontology
modularization is the possibility of knowledge
contextualization (different parts of the ontology
may correspond to different contexts) and
knowledge personalization, i.e. ownership and
authorization (Parent and Spaccapietra, 2009).
Several different approaches/techniques can be
found on the literature, varying in terms of
requirements and intents. In this paper, we are only
interested in three distinct kind of approaches: (i)
Ontology Partitioning (Del Vescovo et al., 2011;
Stuckenschmidt and Schlicht, 2009), (ii) Module
Extraction (Seidenberg, 2009; Hussain and Abidi,
2010), and (iii) Ontology Summarization (Peroni et
al., 2008; Zhang et al., 2009; Li et al., 2010). Next
we describe each of these approaches. According to
(d’ Aquin et al., 2009), Ontology Partitioning is seen
as the task of splitting up an ontology (cf. Definition
1) into a set of (probably disjoint) modules (cf.
Definition 2) such that the union of all the resulting
modules is semantically equivalent to the original
ontology.
Definition 1 (Ontology) – An ontology (also
known as knowledge base) is a tuple
,
where is the terminological axioms and is the
assertional axioms. Both are defined based on a
structured vocabulary
,
comprised of
concepts and roles . Concepts (and roles) axioms
are of the form ⊑ (⊑) or ≡ (≡)
such that ,∈ (,∈) respectively. For a set
of individuals , concepts and roles assertions are of
form
or
,
such that ∈, ∈ and
,,∈.
The semantics related to an ontology is provided
by an interpretation
over a domain Δ such that it
maps: (i) the elements of the domain to the ontology
instances, (ii) the subsets of the domain to the
ontology concepts, and (iii) the binary relations on
the domain to the ontology roles.
An ontology partitioning identifies the key topics
of an ontology and splits it into several fragments
(Stuckenschmidt and Schlicht, 2009). Typically,
each key topic gives rise to a fragment which is
usually called as module (cf. Definition 2).
Definition 2 (Module) – A module of an
ontology
,
is defined as
,
, where
⊆ and
⊆ are the axioms
dealing with (i) concepts
, (ii) roles
and (iii)
individuals
such that: (a)
⊆, (b)
⊆ and
(c)
⊆ respectively. Accordingly, an ontology
KEOD2012-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
266
module is per se an ontology too.
Ontology partitioning is formalized as follows.
Definition 3 (Ontology Partitioning) – The
Partitioning task is seen as a function :→
where an ontology is splitted into a set of modules
with elements (modules) such that
,
,…,
.
A module extraction aims to extract a focused
fragment (or module) of the original ontology given
a specific topic of interest (Hussain & Abidi 2010).
The topic of interest is captured by the notion of
signature (cf. Definition 4).
Definition 4 (Signature) – A signature to extract
a module
,
from
,
is
defined as
,
where
⊆
⊆
and
⊆
⊆ are the axioms (concepts
,
roles
and individuals
) specifying the context
of the module to be extracted such that:
⊆
⊆
,
⊆
⊆ and
⊆
⊆.
A module extraction is formalized as follows.
Definition 5 (Module Extraction) – The Module
Extraction task is seen as a function :
,
→
where an ontology module is extracted from an
ontology according to a given signature .
Ontology summarization provides a succinct
representation (or compressed version) of the
ontology (referred to as summary) emphasizing the
topics contained in an ontology according to
visualization and navigation purposes (Zhang et al.
2009; Li et al. 2010).
Definition 6 (Summary) – A summary description
of an ontology
,
is defined as
,
where
⊆ and
⊆ are the axioms
specifying the concepts
, the roles
and the
individuals
that summarize the ontology such that:
⊆,
⊆ and
⊆ respectively.
Ontology summarization is then formalized.
Definition 7 (Ontology Summarization) – The
Ontology Summarization task is seen as a function
:→ where a description is generated to
summarize the ontology .
It is worth notice that from the perspective of an
ontology, the notions of (i) module (), (ii)
signature ( and (iii) summary () have similar
formal definitions. However, these notions differ on
their purpose and in extension (in terms of set
inclusion), such that:
⊆⊆
⊆⊆
No relation can be defined between and .
4 I3OM PROCESS
The I3OM’ proposal presented here is an assisting
tool for iteratively, incrementally, and interactively
navigate and retrieve information from repositories
described by ontologies. We argue that the
combination of ontology modularization techniques
into an iterative, interactive and incremental process
helps the users perceiving the original knowledge
base by reducing its complexity and size. The
approach is novel in several aspects:
Iterative, because the process phases are
repeated several times (iterations);
Incremental, because the result is being
progressively built/refined along the iterations;
Interactive, because the user is requested to
participate in the process by refining/indicating
the navigation direction;
Semantic-based, because the process relies on
and is driven by the -box underlying the data
-box;
The refinement process is not a progressive
intersection of terminological terms (e.g.
concepts), but instead is a signature-based
ontology modularization whose modules are
not disjoint in any iteration.
Algorithm 1 captures the I3OM approach. The
process is comprised by two distinct steps: Step 1
(line 1 to 6) and Step 2 (line 7 to 12).
In Step 1, the algorithm starts by splitting into
a set of modules by the Ontology Partitioning
which ensures that all the knowledge of the
original ontology is preserved in the respective
modules and is recovered by joining all the modules
(Del Vescovo et al. 2011). Afterwards, a
summarization algorithm is applied to each module
∈. Each resulting summary
contains the
main topics of the extracted module. Consequently,
the set of the resulting summaries contains the
main topics of the original ontology organized by
modules. A conservative ontology summarization
algorithm is required in order to guarantee that every
ontology entity is reachable through . Accordingly,
in each iteration of the I3OM process, provides a
global view of the ontology that allows the further
specification and/or refinement of the query upon .
I3OM-AnIterative,IncrementalandInteractiveApproachforOntologyNavigationbasedonOntologyModularization
267
Algorithm 1: I3OM
Require: An ontology and a signature
Ensure: A summary of the relevant ontology module
is provided together with a set of complementary
summaries .
1:
2:
∅
3:
for all
∈ do
4:
5:
∪
6:
end for
7:
do
8:
,
9:
10:
,
,
11:
while ()
12:
return
,
Next, the algorithm extracts a module from the
ontology and summarizes it (Step 2). In line 8 a
contextualized module
is extracted from the
entire ontology with regard to a given signature
(provided by the user) through ,. Yet, since
the current signature may contain axioms belonging
to several of the initial ontology modules , the
resulting contextualized module may not correspond
to any module in . Instead, the resulting
contextualized module may be subsumed by:
a single initial module (
⊑
∈); or
the union of multiple initial modules (
⊑
∀:
∈⊑).
The contextualized module
is further
summarized in order to obtain a contextualized
summary
(line 9). Thus,
represents the
semantic context (i) to the previous user query and
(ii) to the semantic resources to be retrieved as
response to the query.
The function (line 10) represents the
application module that makes use of I3OM, either
automatically or through the user. The input of the
function is the set obtained in Step 1 and
the set
processed in the current iteration. This
function allows the selection of a set of entities ()
to constraint the next iteration according to four
intends:
Constraint focus: it occurs when the user only
selects ontology entities from
and all of them
are subsumed by the ontology entities selected in
the previous iteration;
Expand focus: it occurs when the signature
selection includes ontology entities of previous
iteration and adds new ones existing in
;
Shift focus: it occurs when the selected signature
is comprehended in
;
A combination of the previous three.
In any of these cases, takes the Boolean
value “true”. Alternatively, the function
might decide to stop the I3OM process. In such case,
takes the Boolean value of “false”.
5 WALK-THROUGH EXAMPLE
To demonstrate the proposal we present now a real
walk-through example. For that we use the EKAW
ontology (EKAW 2011) that has a
Description Logics expressivity and it is composed
by 74 concepts, 33 object properties, and it has no
data properties and individuals.
In Step 1, the ontology is split into four modules
,
,
,
. Each one of these modules
is further summarized such that
,
,
,
. Table 1 and Table 2 illustrates
the obtained results. These results do not change
along the iterations (Step 1 is performed once).
Table 1: Metrics of the modules obtained in Step 1.
No. of Classes
56 4 34 5
No. of Properties
30 2 11 2
Table 2: Summaries obtained in Step 1.
∈
∈
,,,
_
,
,
_
,
_
_
,
,
_
Next, Step 2 is performed for the first time (iteration
1). In this iteration the input signature is empty
(
∅). Consequently, the contextualized module
and its summary are also empty (
∅).
According to the output of the function
(
in Table 3), Step 2 runs from iterations 2 to 5.
The input to the function provided by
Step 2 (
) is also depicted in Table 3 together with
few characteristics of the contextualized module
from which
is obtained.
Second iteration starts by the feeding
the I3OM algorithm with
. Considering
a new
contextualized module is extracted and summarized
as
. Considering and
the returns
Ste
p
1
Ste
p
2
KEOD2012-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
268
Table 3: Characteristics of the Extracted Contextualized Modules and its Summary.
It.
Input Information
No. of
Concepts
No. of
Properties
1
∅
0 0
∅
2
31 10
,,
_
,
_
3
20 8
,,
_
,
_
,
4
_
,
34 12
,,,
_
,,
_
5
6 0
,
_
which contains only an entity () of
.
Therefore, the is constraining the focus of
the relevant information. This is further confirmed
by the characteristic of the extracted contextualized
module as well by its summary (
). Next,
considering and
the returns
which
contains an entity (_) of
and
another entity () of
, which suggests
that the is expanding its focus. This
suggestion is confirmed by the characteristic of the
extracted contextualized module as well by its
summary (
). Finally, considering and
, the
returns
which contains no entities of
, but only an entity () of
,
which means the is shifting its focus. This
is proved by the resulting contextualized module and
its summary (
).
This real walk-through example demonstrated
the capabilities and effectiveness of the I3OM
process in supporting the different intends of
querying/retrieving information: constraining,
expanding and shifting.
6 RELATED WORK
The KC-Viz (Motta et al. 2011) is a plugin for the
Neon Toolkit (Neon Toolkit 2012) that enables the
user to visualize and navigate through ontologies.
This approach exploits the Key Concepts Extraction
(KCE) (Peroni et al. 2008) ontology summarization
algorithm to identify concise overviews of the
ontology and support the ontology navigation
starting from the most useful concepts for making
sense of an ontology. This is enhanced by a
powerful user interface comprehending a panoply of
graphical features (e.g. zooming, layout
customization) (Motta et al. 2011).
However, while KC-Viz supports ontology
navigation, it does not allow the user to focus on a
particular set of entities and its related entities (i.e. a
contextualized module). On the contrary, our
approach enables the user to focus on an ontology
module according to a set of selected entities.
Moreover, KC-Viz navigation is carried through a
tree-structure, which only reflects the subsumption
relations. Therefore, it (i) only allows the user to
focus on the sub-classes of a node, and (ii) it does
not capture other types of relations. As our approach
relies on the Module Extraction task in each
iteration, all relations are always available to the
user/application. Additionally, while all the ontology
is reachable in every iteration of I3OM, in KC-Viz
this is not always true. Yet, the powerful user
interface features of KC-Viz are useful and can be
exploited by the I3OM’ function.
7 SUMMARY AND DISCUSSION
The proposed ontology modularization-based
process benefits from the advantages that each
particular modularization technique has. While
splitting the ontology into smaller modules that
emphasize the topicality of the ontology and
enhance the visualization, Ontology Partitioning
guarantees that all the knowledge of the original
ontology is preserved (d’ Aquin et al. 2009).
Ontology Summarization has the ability to extract
the key entities out of the ontology which may
represent the key areas covered by the ontology.
Module Extraction extracts specialized knowledge
from different topics according to a signature. This
signature is indeed a core concept in the proposal as
it allows the interaction between the application/user
and the automatic process in a stateless way.
Preliminary experiments with the I3OM
prototype showed that users are able to easily,
I3OM-AnIterative,IncrementalandInteractiveApproachforOntologyNavigationbasedonOntologyModularization
269
efficiently and effectively navigate through the
ontology, reaching their goal in a small number of
iterations. Further, the more the users are proficient
with a search approach (text-based search vs.
ontology-based search), the fast they answer the
questions and less intellectual effort they put on the
task. Observations showed that the time spent to
answer a question with the I3OM system decreased
in the latter questions despite these questions were
not simpler than the earlier ones. Moreover, medium
and high-proficient users expressed their sympathy
for the I3OM approach, while answering the
questions faster with the IO3M system. However,
these experiments also demonstrate that the
combination of third-party ontology modularization
algorithms into the I3OM process is not trivial and
demands significant improvements in order to deal
with ontologies having disparate set of
characteristics. Therefore, this issue is requiring our
current and future attention.
Another identified major issue, which is not
directly related to the I3OM process but, instead, is
related to the World Search project overall approach
concerns the GUI module. In fact the users
expressed concerns about the supplied GUI,
suggesting the need to better track the
results/iteration. In this respect, the GUI must
automatically adapt (change based on several factors
such as (i) the user proficiency, (ii) the content’
complexity of the provided semantic context (e.g.
shown by means of a tree or a graph) and (iii)
provide specific interaction for the orthogonal
ontological dimensions (e.g. time and space).
ACKNOWLEDGEMENTS
This work is partially supported by the Portuguese
projects: World Search (QREN11495) and OOBIAN
(QREN 12677), both funded by FEDER through the
COMPETE program for operational factors of
competitivity.
REFERENCES
Berners-Lee, T., Hendler, J. & Lassila, O., 2001. The
Semantic Web. Scientific American, pp.29-37.
d’ Aquin, M. et al., 2009. Criteria and Evaluation for
Ontology Modularization Techniques. In Modular
Ontologies. Springer, pp 67-89.
Del Vescovo, C. et al., 2011. The modular structure of an
ontology: atomic decomposition. In Fifth Int.
Workshop on Modular Ontologies. IOS Press.
Dzbor, M. et al., 2006. Developing Ontologies in OWL:
an Observational Study. In Workshop on OWL:
Experiences and Directions. Vol-216 CEUR.
EKAW. EKAW Ontology. [Online] Available: http://oaei.
ontologymatching.org/2011/conference/data/ekaw.owl
Franconi, E., Guagliardo, P. & Trevisan, M., 2010. An
intelligent query interface based on ontology
navigation. In Workshop on Visual Interfaces to the
Social and Semantic Web.
Hussain, S. & Abidi, S. S. R., 2010. Extracting and
merging contextualized ontology modules. In Fourth
Int. Workshop conference on Modular Ontologies.
IOS Press, pp 25-40.
Li, N., Motta, E. & d’ Aquin, M., 2010. Ontology
summarization: an analysis and an evaluation. In Int.
Workshop on Evaluation of Semantic Technologies.
Shangai, China.
Motta, E. et al., 2011. A novel approach to visualizing and
navigating ontologies. In Tenth Int. Semantic Web
Conference, pp.470-486.
Neon Toolkit 2012. [Online] Available: http:// neon-
toolkit.org.
Parent, C. & Spaccapietra, S., 2009. An Overview of
Modularity. In Stuckenschmidt, C. Parent, & S.
Spaccapietra, eds. Modular Ontologies. Springer, pp.
5-23.
Peroni, S., Motta, E. & d’ Aquin, M., 2008. Identifying
key concepts in an ontology, through the integration of
cognitive principles with statistical and topological
measures. In Seventh Int. Semantic Web Conference,
pp. 242-256.
Seidenberg, J., 2009. Web Ontology Segmentation:
Extraction, Transformation, Evaluation. In Heiner
Stuckenschmidt, C. Parent, & S. Spaccapietra, eds.
Modular Ontologies. Springer, pp 211-243.
Stuckenschmidt, H. & Schlicht, A., 2009. Structure-Based
Partitioning of Large Ontologies. In Modular
Ontologies. Springer, pp. 187-210.
WS, 2009. [Online] Available: http://www.microsoft.
com/portugal/mldc/worldsearch.
Zhang, X. et al., 2009. Summarizing vocabularies in the
global semantic web. Journal of Computer Science
and Technology, 24(1), pp.165-174.
KEOD2012-InternationalConferenceonKnowledgeEngineeringandOntologyDevelopment
270
