KNOWLEDGE MANAGEMENT AND ACQUISITION

IN DIGITAL REPOSITORIES

A Semantic Web Perspective

Dimitrios A. Koutsomitropoulos, Georgia D. Solomou, Andreas D. Alexopoulos

and Theodore S. Papatheodorou

High Performance Information Systems Laboratory, School of Engineering, Dpt. of Computer Engineering

and Informatics, University of Patras, Building B, 26500, Patras-Rio, Greece

Keywords: Knowledge Management, Metadata, Interoperability, Ontologies, Reasoning, Digital Repositories.

Abstract: Information management, description and discovery, as they are today implemented in digital repositories

and digital libraries systems, can surely benefit from the stack of Semantic Web technologies. Most

importantly, the ability to infer implied information over declared facts and assertions, based on their rich

descriptions and associations, can span new possibilities in how stored assets can be accessed, searched and

discovered. In this paper we propose a process and implementation that provides for inference-based

knowledge discovery, retrieval and navigation on top of digital repositories, based on existing metadata and

other semi-structured information. We show that it is possible to produce added-value and meaningful

results even when existing descriptions are only flatly organized and we achieve this with little manual

intervention. Our work and results are based on real-world data and applied on the official University of

Patras institutional repository that is based on DSpace.

1 INTRODUCTION

The Semantic Web infrastructure mainly relies on

specifications for expressing ontologies in web-

compatible format, like OWL or OWL 2 (Grau, et

al., 2008) and, lately, on programming efforts for

manipulating such ontologies, like the OWL API

(Horridge, et al., 2007) and the Protégé 4.0 code-

base (the CO-ODE project, http://www.co-ode.org/).

To be able to fully reap the benefits of a Semantic

Web, reasoning over ontological information is of

exceeding importance, a fact that was sometimes

overlooked in the past, possibly because of the

immaturity of available tools and techniques. To our

belief, the ability to infer implied information over

declared facts and assertions is one of the most

prominent reasons to investigate and implement the

Semantic Web, in a sense that adds an “AI” flavor to

the current Web (Hendler, 2008).

Therefore, in this paper we present and document

a process that builds upon the well-known digital

repositories paradigm and enhances it with the

Semantic Web‟s features. The main goal that drives

our efforts is not to re-implement a digital repository

system using Semantic Web APIs and technologies,

but to provide inference-based knowledge discovery,

retrieval and navigation on top of such a system,

based on existing metadata and other semi-

structured information.

To prove our concept, we describe a concrete,

working prototype that provides for inference-based

search and navigation on top of the DSpace digital

repository system. DSpace has become a popular

open-source digital repository solution with one of

the most rapidly growing user bases worldwide.

DSpace metadata follow the Dublin Core (DC)

specification by default, while it is possible to

import and use other metadata schemata as well.

This paper is further organized as follows: First

an overview of related work on digital libraries and

semantics is given; then, we introduce the process

for constructing the repository‟s ontology and point

out its most important aspects. Following, we

document our extensions to the DSpace system and

the implementation of the ontology management,

search, navigation and reasoning services. Finally

we give specific examples and results that

demonstrate these new capabilities and summarize

the conclusions of our work. A partial description of

117

Koutsomitropoulos D., Solomou G., Alexopoulos A. and Papatheodorou T. (2009).

KNOWLEDGE MANAGEMENT AND ACQUISITION IN DIGITAL REPOSITORIES - A Semantic Web Perspective.

In Proceedings of the International Conference on Knowledge Management and Information Sharing, pages 117-122

 SciTePress

this work and source code are freely available at:

http://wiki.dspace.org/ index.php/User:Kotsomit.

2 RELATED WORK

Some widely adopted mechanisms for digital

repositories that, similar to our work, appear to

utilize Semantic Web technologies are BRICKS,

SIMILE, Fedora and JeromeDL plus the more

recently appeared Talia.

Both BRICKS (Risse, et al., 2005) and Fedora

(http://www.fedora.info) provide the basic

architecture upon which digital library applications

can be developed and deployed. Their functionality

is further enhanced by supporting some basic

Semantic Web technologies, like the expression of

relations between objects in RDF and the retrieval of

data through the evaluation of queries using

SPARQL or other RDF query languages.

Through the application of RDF and Semantic

Web techniques, SIMILE (http://simile.mit.edu/)

offers DSpace improved support for arbitrary

schemata and metadata and provides an architecture

for disseminating digital assets. Among SIMILE‟s

implemented tools, the one that pertains mostly to

our work is a faceted browser, known as Longwell

(its DSpace version is called Dwell), that gives the

ability to cross-section the data along fixed

dimensions of structured metadata. Furthermore,

SIMILE provides tools to merge lexical or semantic

variants via simple inferencing. Nevertheless, the

current implementation of SIMILE does not seem to

offer reasoning based querying.

JeromeDL (Kruk, et al., 2005) is a “social

semantic digital library” that stores its metadata in

RDF all along, by utilizing a corresponding RDF

store (Sesame). A level of inference is supported

through a simple recommendation engine based on

Prolog. JeromeDL seems to solve the “semantic

bootstrapping” problem following a bottom-up

approach, since the ontological schema is

constructed and populated in advance. In such a way

the problem of retrieving semantic implications and

inference-based results, also from flatly organized

relational data base sources, is circumvented.

Finally Talia (Nucci, et al., 2008) is a library

platform that stores its metadata into a relational DB

schema and keeps it “in sync” with an RDF data

store (Redland). In addition, it provides a unified

query interface for both database and RDF metadata

using SQL or SPARQL, as required. However, Talia

does not do any inferencing on the RDF data and

leaves this responsibility to the underlying RDF

store.

3 CREATION AND POPULATION

OF THE ONTOLOGICAL

MODEL

In this section we give a brief outline of how we

have developed a Semantic Web ontology out of the

repository‟s metadata, based on which we can

employ our complementary, semantics-aware

services. A more detailed description of this process

is out of the scope of this paper and can be sought in

(Koutsomitropoulos, et. al., 2008b, 2009b).

Based on the DC RDF(S) schema (Nilsson, et al.,

2008) we have developed a semantic application

profile (Koutsomitropoulos, et al., 2009a) in three

main steps:

 First, we transferred the DC original schema in

OWL format.

 Then we augmented its semantics, by using

property characteristics not available in

RDFS: for example, we have identified some

DC properties to be inverse, symmetric or

transitive and declared them as such.

 We further profiled the model by including

refinements for our particular application, that

is, the University of Patras digital repository.

We have modelled vocabularies in

taxonomies, introduced new properties for

DSpace relations („author‟, „sponsorship‟) and

new classes for DSpace notions („item‟,

„collection‟) that the original DC does not

provide for. Further, we used OWL 2-specific

constructs, like role-chains, to represent

intrinsic complex relations, like the „co-

author‟ relationship between authors.

A naive attempt to model the DC domain as

thorough as possible, by representing each and every

potential semantic relationship, can easily render the

ontology undecidable (Koutsomitropoulos, et al.,

2008b). However we have identified that the

punning feature, introduced with OWL 2, can

reasonably deal with ambiguities and meta-

modelling requirements, inherent in the DC

specification.

The resulting ontology, including the new

refinements, is then populated in an automated way

from metadata already existing within the live

DSpace installation of the University of Patras

institutional repository. These metadata are

harvested through the repository‟s OAI-PMH

interface (Lagoze, et al., 2002) and mapped to the

ontology using an XSLT developed for this purpose.

KMIS 2009 - International Conference on Knowledge Management and Information Sharing

118

4 SEMANTIC SEARCH

AND NAVIGATION

In this section we discuss the design decisions and

the implementation of the semantic enhancements to

the DSpace digital repository system.

4.1 Design Goals

Most of the design decisions stem from a set of

requirements that where posed beforehand, in order

to guarantee reproducibility and applicability of our

efforts, as well as to ensure the potential of the

semantic services offered. They can be summarized

as follows:

Interoperability. At its core, interoperability is

achieved by adhering to information standards, at

any level: Repository‟s metadata are structured

according to the XML format, ontology models are

represented using W3C‟s OWL and OWL 2

specifications and the semantic gap between the two

is bridged using an XSLT transformation. Further,

this transformation upgrades interoperability to a

semantic level, by rendering the repository‟s

metadata descriptions semantically compatible to the

ontology‟s structures. Lacking a communication

protocol for the exchange of OWL information, we

at least opt for wrapping and transforming OAI-

PMH responses that offer a standard way for

harvesting metadata from data providers. This comes

in contrast to accessing the repository‟s database

directly, as this could be dependent on the

proprietary DB schema. As a result, our approach

could be seamlessly integrated with other resource

management systems, at least OAI compliant ones.

Support for OWL 2. The need to support this

newly proposed extension to OWL comes from the

fact that OWL 2 is able to represent a richer set of

semantics than its predecessor, thus enabling more

advanced inferences. On the other hand, this reduces

our choices of inference engines to only supporting

ones, such as FaCT++ and Pellet. Performance is not

our main concern here, since the OWL 2 reasoning

algorithm is known to be scalable (Horrocks, et al.,

2006) and its implementation in these reasoners is

heavily optimized. However, we are forced to use a

direct in-memory implementation, since none of

these reasoners supports a communication interface

other than DIG, which is currently incompatible

with OWL DL, not mentioning OWL 2

(Koutsomitropoulos, et al., 2008a). Therefore, high

expressivity comes at the cost of a truly distributed

3-tier architecture.

Extensibility. A major design decision was to

totally implement our extensions using the OWL

API, while avoiding references to DSpace specific

methods and classes. OWL API equips us with a

satisfactory layer of abstraction on top of which

further extensions can be implemented. In addition,

it does not restrict us to any particular inference

engine or a specific reasoning approach: The

selection of the reasoner class constructed can be

easily parameterized, while the reasoning strategy

can stay the same, rendering our implementation

reasoner-independent. Furthermore, it is easy to

support other querying protocols and/or methods: In

our implementation, a query is given in the form of a

Manchester Syntax class expression (Horridge &

Patel-Schneider, 2008). Just as easily, query

formulation can be extended to follow another

paradigm, such as SPARQL/OWL (Sirin & Parsia,

2007), as soon as its specification grows mature and

a supporting parser is implemented.

Figure 1: Semantic Search interface and the auto-complete

facility.

User-friendliness/Intuitiveness. Unfortunately, a

common way for querying OWL knowledge bases

has not been standardized yet. SPARQL is a

language for querying RDF graphs, but it does not

take into account OWL‟s richer semantics. To build

a “user-friendly” service and keep the complexity of

query formulation as low as possible, we

implemented a query interface based on class

construction using Manchester Syntax, in a way

inspired by the DL Query Tab of Protégé. Query

results are formed by all individuals that are inferred

to be instances of the constructed class. Manchester

Syntax has the advantage to offer a pseudo-natural

English language expression of classes, thus

facilitating, to some extent, the end user to formulate

a query. In addition, we have tried to make this

process intuitive, by implementing an AJAX-based

suggestion and auto-complete mechanism, where

matching entities names are suggested to the user, as

the query is typed in (Figure 1).

KNOWLEDGE MANAGEMENT AND ACQUISITION IN DIGITAL REPOSITORIES - A Semantic Web Perspective

119

Figure 2: Architectural model of the Semantic Search extension of DSpace.

4.2 Architectural Overview

An overview of the services we have built around

DSpace is depicted in Figure 2. The most important

modules and interfaces that enable semantic services

in our digital repository are the following:

 Semantic Search interface, which, in

collaboration with the appropriate inference

engine, allows for the construction,

submission and evaluation of a semantic

query. Retrieved results are displayed here in

the form of a list.

 Semantic Navigation interface is where detailed

ontological information about a selected entity

(individual) is presented.

 Ontology Population refers to the dynamic

construction of the ontology, which comes

from DSpace‟s OAI harvested metadata, after

applying the appropriate XSLT transformation

on them.

 The Inference Engine is responsible for

processing the ontological documents and for

performing reasoning over them. We have

chosen FaCT++ but any other DL reasoner

may be used, as stated in section 4.1.

Context with DSpace itself is indirectly

maintained, since it is still possible to open

DSpace‟s simple item view page from within the

navigation pane (Figure 3).

5 EVALUATION AND EXAMPLES

In this section we give some examples of how

semantic-enabled search and navigation can work

towards discovering and acquiring new and implied

knowledge. This knowledge is impossible to be

retrieved through a traditional querying interface, as

there is not even a way to express such requests

using solely combinations of matching keywords, let

alone reasoning and inference themselves (Horrocks,

2008). Further, we see that these services allow

retrieval and presentation of entities of any type, not

just items.

5.1 Entity Retrieval

In DSpace, the main information unit is the item,

which represents a specific resource (document,

image or other) that has been uploaded in DSpace,

as well as its containers, namely collection and

community. DSpace search, therefore, is targeted

towards retrieval of items only, i.e. search results are

always a list of items or collection and community

names.

In Figure 3, we notice that the item 1987/117 has

a dcterms:type „Book‟. Clicking on „Book‟ we

now see detailed information regarding „Book‟ as an

entity itself. We also notice that we have indirectly

retrieved every item that has type „Book‟ (through

the inverse dcterms:type property). In addition,

we find out that „Book‟ belongs to the dspace-

ont:dspacetype class, clicking on which we

trigger semantic search to fetch all instances of this

class, which of course are not DSpace items. The

same naturally holds for other indirect DSpace

entities we have reified, such as authors, formats etc.

5.2 Improved Search Results and

Knowledge Discovery

Another advantage of semantic search is to allow

retrieval of more as well as more precise results.

These results may be implied by the current data

model, but there is no way to retrieve them using the

standard configuration.

KMIS 2009 - International Conference on Knowledge Management and Information Sharing

120

Figure 3: Individual view and navigation pane for item 1987/117.

As an example, suppose we would like to

retrieve all items that contain image files. Searching

with the keyword „image‟ in the traditional

repository search returns 2 items. However,

examining each of these items metadata reveals that

only the latter has actually an „image/gif‟ format; the

other has a format of „application/pdf‟. The reason

why it is returned by traditional search is that

DSpace searches also inside the document‟s text and

happens to meet the word „image‟.

On the other hand, the corresponding query

through the semantic search interface (Query 1 in

Table 1) fetches just one item, exactly the one that

has format „gif‟. In this sense, a better level of query

precision could be sought, since more precise and

semantically accurate results can now be obtained.

Suppose now we would like to find out who

draws sponsorship from a specific institution, for

example, what is funded by the „Hellenic Ministry of

Culture‟ (Query 2). Searching with these keywords

through traditional search returns an item that

includes this organization‟s name in its

„sponsorship‟ metadata field.

Semantic search however retrieves also the

author of this item, aside from the item itself. This is

a direct consequence of a role-chain we have

declared in our ontology, conveying the fact that

authors of items are also receiving sponsorship from

the same institution. This example then suggests

how semantic search could also improve the recall

of retrieval, by obtaining a greater number of results.

Table 1: Example queries using the semantic search

interface.

Query (in Manchester Syntax)

Ask for:

dcterms:format some dspace-

ont:image

Items that contain

image files

dspace-ont:sponsorship value

dspace-ont:

Hellenic_Ministry_of_Culture

Items/authors that

draw sponsorship

from a specific

institution

inv(dspace-ont:author) some

(dcterms:format min 2

owl:Thing)

Authors of items that

have at least two

different formats

dspace-ont:co_author some

(foaf:name value “Bekiari”)

The co-authors of an

author

In the „image‟ example above, semantic search is

able to fetch the particular item, despite the fact that

its format is declared just as „gif‟ (i.e. it does not

contain the keyword „image‟). This is because in our

DSpace‟s „Item View‟ page for item 1987/117

Members of the „dspacetype‟ class

Ontological info about the „Book‟ individual

KNOWLEDGE MANAGEMENT AND ACQUISITION IN DIGITAL REPOSITORIES - A Semantic Web Perspective

121

ontology, „gif‟ is an instance of the „image‟ class

and thus the underlying reasoner is able to conduct

an inference; that is, since we ask for an „image‟

format, we also ask for every instance of this class.

This knowledge discovery capability can also be

determined by asking, for example, for the authors

of those items (Query 3) that have at least two

different formats, using a cardinality restriction on

dcterms:format. It is easy to see that such a query

is impossible to be expressed through traditional

search. Similarly, with Query 4 we ask for the co-

authors of an author, based on her surname. Due to

the definition of the co_author property as a role-

chain, this request becomes possible and the result is

straightforward.

6 CONCLUSIONS

In this paper we have shown how to augment

traditional digital repository services by

implementing an extensible semantic search and

navigation facility on top of DSpace. This facility

relies purposely on the OWL API and is designed to

be independent of the underlying system, following

a “plug-in” philosophy. In combination with the

ontology creation and population process, this

facility could semantically enable any web-based

digital repository system.

Our results confirm that it is possible to navigate

among a repository‟s metadata in more flexible and

associative ways. In addition, semantic search can

improve traditional keyword search by retrieving

more items, but also by fetching more semantically

accurate results. And of course, semantic search

allows the expression of queries that cannot be

expressed by simple keyword-based retrieval.

Finally, it can be seen that the use of ontologies

in digital repositories and other information systems,

in the way it is suggested in this paper, can benefit

from an ontology harvesting and exchange protocol

(just as OAI does for metadata), as well as from a

standard and semantics-aware language for querying

OWL documents.

REFERENCES

Grau, B.C. et al., 2008. OWL 2: The next step for OWL.

Web Semantics: Science, Services and Agents on the

World Wide Web, 6(4), pp. 309-322.

Hendler, J., 2008. Web 3.0: Chicken Farms on the

Semantic Web. Computer, 41(1), pp. 106-108.

Horridge, M., Bechhofer, S. & Noppens, O., 2007.

Igniting the OWL 1.1 Touch Paper: The OWL API. In

OWLED 2007, 4th International Workshop on OWL

Experiences and Directions.

Horridge, M. & Patel-Schneider, P., 2008. Manchester

Syntax for OWL 1.1. In OWLED 2008, 5th Int.

Workshop on OWL Experiences and Directions.

Horrocks, I, Kutz, O. & Sattler, U., 2006. The Even More

Irresistible SROIQ. In KR2006, 10th Int. Conf. on

Principles of Knowledge Representation and

Reasoning.

Horrocks, I., 2008. Ontologies and the Semantic Web.

Communications of the ACM, 51(12), pp. 58-67.

Koutsomitropoulos, D., Meidanis, D., Kandili, A. &

Papatheodorou, T., 2008a. Establishing the Semantic

Web Reasoning Infrastructure on Description Logic

Inference Engines. In Y. Manolopoulos et al., eds.

Enterprise Information Systems, Lecture Notes in

Business Information Processing. Springer, pp. 351-

362.

Koutsomitropoulos, D., Solomou, G. & Papatheodorou,

T., 2008b. Semantic Interoperability of Dublin Core

Metadata in Digital Repositories. In Innovations’08,

5th Int. Conf. on Innovations in Information

Technology.

Koutsomitropoulos, D., Paloukis, G. & Papatheodorou, T.,

2009a. Semantic Application Profiles: A Means to

Enhance Knowledge Discovery in Domain Metadata

Models. In M.A. Sicilia and M. Lytras, eds. Metadata

and Semantics. Springer, pp. 23-34.

Koutsomitropoulos, D., Solomou, G., Alexopoulos, A. &

Papatheodorou, T., 2009b (in press). Semantic

Metadata Interoperability and Inference-Based

Querying in Digital Repositories. Journal of

Information Technology Research. (Accepted for

publication March 2009).

Kruk, S.R., Decker, S. & Zieborak, L., 2005. JeromeDL -

Reconnecting Digital Libraries and the Semantic Web.

In WWW2005, 14th Int. World Wide Web Conference.

Lagoze, C., Van de Sompel, H., Nelson, M. & Warner, S.,

2002. The Open Archive Initiative Protocol for

Metadata Harvesting. [Online] Available at:

http://www.openarchives.org/OAI/openarchivesprotoc

ol.html [Accessed December 2008].

Nilsson, M, Powell, A., Johnston, P. & Naeve, A., 2008.

Expressing Dublin Core metadata using the Resource

Description Framework (RDF). DCMI

Recommendation. [Online] Available at:

http://dublincore.org/documents/dc-rdf/ [Accessed

January 2009]

Nucci, M., Hahn, D. & Barbera, M., 2008. The Talia

Library Platform - Rapidly Building a Digital Library

on Rails. In SFSW2008, 4th Workshop on Scripting for

the Semantic Web, European Semantic Web

Conference.

Risse, T., Knezevic, P., Meghini, C., Hecht, R. & Basile,

F., 2005. The BRICKS Infrastructure - An Overview.

In EVA 2005, 8

Annual Int. Conf. on Electronic

Information and Visual Arts.

Sirin, E. & Parsia, B., 2007. SPARQL-DL: SPARQL

Query for OWL-DL. In OWLED 2007, 4th Int.

Workshop on OWL Experiences and Directions.

KMIS 2009 - International Conference on Knowledge Management and Information Sharing

122