A Semantic Framework to Enrich Collaborative Tables with Domain
Knowledge
Anna Goy, Diego Magro, Giovanna Petrone, Marco Rovera and Marino Segnan
Dipartimento di Informatica, Università di Torino, C. Svizzera 185, 10149, Torino, Italy
Keywords: Communication, Collaboration and Information Sharing, Metadata Management, Linked Open Data,
Semantic Web, Ontology-driven Applications.
Abstract: In this paper we present a project aimed at enhancing a collaborative environment for resource management
(SemT++) with domain knowledge, represented by a local ontology and a connection to external data,
retrieved from Linked Open Data sets. Our approach is based on the assumption that heterogeneous
resources can be viewed as "information objects", and can be organized within collaborative spaces (i.e.,
"round tables"). Information objects, among other properties, are characterized by their content. Annotations
representing resource content (e.g., "Torino") can thus be linked to domain knowledge which provides users
with useful information. We tested this approach on the geographic domain, by connecting resources to
commonsense geographic knowledge and to information available in GeoNames.
1 INTRODUCTION
In the current ICT scenario, Human-Computer
Interaction (HCI) and Personal Information
Management (PIM) (Barreau and Nardi, 1995) have
to face new challenges. In particular, new web
architectures and paradigms, such as Web 2.0, Cloud
Computing, Software-as-a-Service, are posing new
problems and offering new opportunities. Two
aspects are particularly relevant from our viewpoint:
first, users have to face the management of a huge
amount of heterogeneous resources, possibly related
to the same content, but encoded in different
formats, handled by different applications, stored in
different places, and belonging to different types
(documents, emails, videos, bookmarks,...); second,
users can actively participate in content creation, can
share resources and knowledge, and can collaborate
with each other in carrying on many activities.
One of the possible approaches to effectively
support both heterogeneous resource management
and collaboration relies on semantic technologies,
which can be exploited to provide users with smarter
and more friendly tools for managing shared
resources on the web. This idea is not new. In
particular, a significative trend in this direction is the
emerging Social Semantic Web (Breslin et al., 2009),
which relies on the idea that semantic technologies
can support the creation of machine readable
interlinked representations of social objects (people,
contents, resources, tags, etc.) enabling different
social "islands" (i.e., isolated communities of users
and data) to be connected and integrated. The
approach presented in this paper can be seen as part
of this project, since it aims at enhancing a
collaborative environment for resource management
with semantics, in order to provide users with a
smarter support to resource management.
Our approach, in particular, is based on the
hypothesis that digital resources should be viewed as
information objects, and should be managed in a
uniform way, independently from their possibly
heterogeneous types. Awareness about information
objects includes different aspects, such as
knowledge about the format they are encoded in
(e.g., PDF, HTML, JPEG, etc.), about their structure
(e.g., if a document contains images or hyperlinks),
and about their content (e.g., what a document "is
about", or what an image represents). This kind of
knowledge has been encoded within Semantic Table
Plus Plus (SemT++), an environment aimed at
supporting users in collaborative resource
management on the web.
In this paper, we describe an enhancement of
SemT++, leading to DSemT++ (Domain-aware
SemT++). Besides general knowledge about
information objects, i.e., information resources as
Goy, A., Magro, D., Petrone, G., Rovera, M. and Segnan, M..
A Semantic Framework to Enrich Collaborative Tables with Domain Knowledge.
In Proceedings of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2015) - Volume 3: KMIS, pages 371-381
ISBN: 978-989-758-158-8
Copyright
c
 2015 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
371

such, DSemT++ relies on knowledge about their
content. The first type of knowledge is "universal",
in the sense that it is (to a certain extent) domain
independent, i.e., it models digital resources
independently of their specific content (for example,
a digital resource always is encoded in a given
format, is expressed in one or more languages, and
so on); knowledge about resource content is usually
domain-specific, since resource content can refer to
very different knowledge domains: if a document
talks about European Medieval history, the semantic
knowledge enabling a tool to deal with the resource
content (e.g., for retrieving it) must include a
semantic representation (e.g., a Domain Ontology)
modeling concepts belonging to the European
Medieval history.
A detailed account of the representation of
knowledge about information objects in SemT++
can be found in (Goy et al., 2014a). In this paper, we
will focus on the second type of knowledge, and we
will show how a Domain Ontology, coupled with
existing resources available as Linked Open Data
sets, can be exploited to support users in the
organization, retrieval and usage of shared digital
resources. The architecture aimed at including
domain knowledge in a resource management
collaborative environment, together with the support
provided by this kind of knowledge, actually
represent the major contribution reported in this
paper.
The rest of the paper is organized as follows. In
Section 2 we set the background, by discussing the
main related work, and in Section 3 we briefly
summarize the SemT++ project, as it is described in
previous papers (mentioned below). In Section 4,
which contains the novel contribution with respect to
our earlier work and represents the core of this
paper, we describe DSemT++, i.e., the enhancement
of SemT++ with domain knowledge; moreover, we
explain why we chose commonsense geographic
knowledge as a testbed domain, we sketch a simple
usage scenario, we describe how domain knowledge
is linked to knowledge in the Linked Open Data
cloud, and how the resulting system supports users
in collaboratively handling semantic descriptions of
digital resources. Section 5 concludes the paper by
discussing open issues and future developments.
2 RELATED WORK
As far as the aspects related to HCI and PIM are
concerned, one of the most relevant research areas is
well accounted for by Kaptelinin and Czerwinski
(2007), which contains a wide presentation of the
problems of the so-called desktop metaphor, and of
the approaches trying to replace it. In particular, one
of the most interesting models discussed in this book
is Haystack (Karger, 2007), a flexible and
personalized system enabling users to define and
manage workspaces referred to specific tasks.
Another interesting family of approaches are those
grounded into Activity-Based Computing − e.g.,
(Bardram, 2007; Voida et al., 2008) − where the
interaction is designed around the concept of user
activity. The main "step forward" of DSemT++ with
respect to these approaches is the explicit domain
knowledge model and the exploitation of Linked
Open Data sets, as explained in Section 4.
Strategies used to organize resources have been
studied also in social tagging systems, where
resources can be tagged with meta-data representing
different aspects (facets), leading to the creation of
folksonomies, i.e., multi-facets classifications
collaboratively and incrementally built by users in a
bottom-up perspective (Breslin et al., 2009).
Interesting improvements of such tagging systems
have been developed by endowing them with
semantic capabilities − e.g., (Abel et al., 2010) − in
particular in the perspective of knowledge workers
(Kim et al., 2009). With respect to these systems,
DSemT++ has a slightly different focus, since it
supports collaboration within (small) groups of
people working together, instead of mass social
communities.
Interesting approaches, based on the definition of
a common conceptual framework provided by
computational ontologies, have been developed
within the Knowledge Management area, with the
aim of facilitating communication and shared
understanding in collaborative decision-making
environments; see, for example, (Evangelou and
Karacapilidis, 2005).
Another important research thread, aiming at
coupling desktop-based user interfaces and Semantic
Web, is represented by the Semantic Desktop
approach (Sauermann et al., 2005). In particular, the
NEPOMUK project (nepomuk.semanticdesktop.org)
defined an open-source framework, based on a set of
ontologies, for implementing semantic desktops,
focusing on the integration of existing applications,
in order to support collaboration among knowledge
workers. (Drăgan et al., 2009) presents an interesting
approach connecting the Semantic Desktop to the
Web of Data, underlying how "connecting the two
networks of information opens up the possibility of
personal services on the desktop which use external
data, but in the personal context of the user, highly
ISE 2015 - Special Session on Information Sharing Environments to Foster Cross-Sectorial and Cross-Border Collaboration between Public
Authorities
372

connected to his personal data and focused on his
interests" (Drăgan et al., 2009, p. 34). Moreover,
"connecting desktop data with the web enables the
system to bring web data to the user, instead of the
user having to go find it by himself" (Drăgan et al.,
2009, p. 35).
This last proposal is one of a large number of
semantic approaches which recently have tried to
exploit the potentiality of the Linked Open Data
(LOD) paradigm, relying on the fact that most
datasets refer to one or more ontologies, or
"semantic" vocabularies (e.g., DBpedia: dbpedia.
org, GeoNames: www.geonames.org). From this
point of view, DSemT++ belongs to the same
research thread.
In the same direction, an interesting project,
which shares many features with our approach, is
LinkZoo (Meimaris et al., 2014) a collaborative
platform which exploits LOD to annotate shared
heterogeneous resources. Semantic descriptions of
resources are stored as RDF triples, and they enable
LinkZoo to couple standard keyword search with
property-based filtering. (Schandl et al., 2012)
contains a survey of the approaches to exploit LOD
in metadata for multimedia content, and CAMO (Hu
et al., 2014) represents an example of linking LOD
to multimedia metadata. Linkify (Yamada et al.,
2014) is an add-on for major browsers which adds a
link to Named Entities recognized in online texts,
pointing to a mashup of information items extracted
from LOD sources. MOAT (Meaning Of A Tag) is a
framework providing a semantic model for defining
machine-readable meanings of tags (Passant and
Laublet, 2008). MOAT models tags as quadruples
(<User, Resource, Tag, Meaning>) and provides a
MOAT server, which can be exploited to share tag
meanings and retrieve them when tagging resources;
in particular, when a user tags content, the MOAT
client retrieves tag meanings from the server and let
the user choose the most relevant one. Tag meanings
are linked to URIs of entities within well-known
LOD datasets, such as DBpedia and GeoNames: this
solves tagging ambiguity (i.e., in case a tag has more
than one URI) and heterogeneity (i.e., in case
different tags refer to the same URI), and enables the
suggestion of relevant content derived from LOD.
Finally, an example of a different exploitation of
LOD can be found in (Giunchiglia et al., 2012),
where the authors present a geospatial ontology,
called Space, based on GeoNames, WordNet and
MultiWordNet, together with the methodology used
for its creation. Space is aimed at representing
geographic and spatial concepts and relations from
the commonsense point of view, an aspect which is
shared by our perspective.
3 THE SEMANTIC TABLE PLUS
PLUS PROJECT
The SemT++ project proposes an interaction model
supporting users in collaboratively handling digital
resources. Such a model is based on the metaphor of
tables, populated by objects, and is described (Goy
et al., 2014b). Tables are thematic contexts, i.e.,
shared workspaces devoted to the management of
specific activities (e.g., the management of a
business project, the organization of children care, a
trip planning). SemT++ tables can be seen as "round
tables", where users can share information and
resources, work together on a document, and so on.
Table participants, in fact, can modify objects, delete
them, or add new ones; invite people to "sit at the
table" (i.e., to become a table participant); define
meta-data, such as comments and annotations.
SemT++ provides an abstract view over objects
lying on tables, by considering them as information
objects that, despite their heterogeneity (they can be
documents, images, to-do items, bookmarks, email
conversations, and so on) can be uniformly
annotated.
Moreover, SemT++ supports workspace
awareness by means of: a table presence panel,
showing the list of table participants currently sitting
at the table; standard awareness techniques, such as
icon highlighting, to notify users about table events
(e.g., an object has been modified); notification
messages, coming from outside SemT++ or from
other tables, filtered on the basis of the topic context
represented by the active table; see (Ardissono et al.
2010).
Figure 1: SemT++ architecture.
Figure 1 shows the relevant components of
SemT++ architecture. The User Interaction
Manager handles all tasks related to the interaction
with users (User Interface generation, and all
communications with the system, namely with the
TO Manager). The TO (Table Object) Manager
plays a "mediation" role between the User
Interaction Manager and the components which
represent the system "intelligence", i.e. the Smart
Object Analyzer and the "semantic" components
(see below). In particular, the TO Manager is in
A Semantic Framework to Enrich Collaborative Tables with Domain Knowledge
373

charge of all the operations which take place on
tables (e.g., adding/deleting objects, comments,
etc.). The Smart Object Analyzer provides the TO
Manager with the analysis of table objects, in order
to discover information about them; for example, it
detects the encoding format (PDF, HTML) and it
looks for parts included in the analyzed object (e.g.,
images, links, etc.). The TO Semantic Knowledge
Manager manages the semantic descriptions of table
objects, which are stored in the TO Semantic KB;
such descriptions are based on the Table Ontology,
which represents the (static) system semantic
knowledge concerning information objects.
The Table Ontology is grounded in the
Knowledge Module of O-CREAM-v2 (Magro and
Goy, 2012), a core reference ontology for the
Customer Relationship Management domain
developed within the framework provided by the
foundational ontology DOLCE (Descriptive
Ontology for Linguistic and Cognitive Engineering)
(Borgo and Masolo, 2009) and some other
ontologies extending it, among which the Ontology
of Information Objects (OIO) (Gangemi et al.,
2005). The Table Ontology enables us to describe
table resources as information objects, with
properties and relations. For example, a table object
can have parts (e.g., images within a document),
which are in turn information objects; it can be
written in English and it can be stored in a PDF file,
or it can be an HTML page; moreover, it has a
content, which usually has a main topic and refers to
a set of entities (i.e., it has several objects of
discourse). Given the object description based on the
Table Ontology, reasoning techniques can be applied
to infer interesting knowledge, mainly from included
parts; for example, if a document contains a
hyperlink to a resource written in French, probably
the document itself is written in French.
A detailed description of this ontology, including
the inferences it enables and how such inferences are
exploited to provide users with suggestions about
object properties can be found in (Goy et al., 2014a).
Within the SemT++ project, we developed a
proof-of-concept prototype, i.e., a Java web
application, deployed on the Google App Engine,
accessible through a web browser. The backend
components, relying on heterogeneous technologies,
are implemented as RESTful Web Services
communicating by exchanging JSON objects. To
store files corresponding to table objects, the current
version exploits Dropbox and Google Drive API,
while Google Mail is used to handle email
conversations. The User Interface (UI) is a dynamic,
responsive single page (client side) application,
exploiting AJAX to exchange JSON objects with a
set of Java Servlets (server-side). UI responsiveness,
guaranteeing immediate availability on different
devices, is supported by Bootstrap
(getbootstrap.com). The Smart Object Analyzer
exploits a Python Parser Service, able to analyze
HTML documents.
Both the Table Ontology and the TO Semantic
KB are expressed in OWL (www.w3.org/TR/owl2-
overview); the TO Semantic Knowledge Manager
exploits the OWL API library
(owlapi.sourceforge.net) to interact with them. The
TO Semantic Knowledge Manager also invokes the
Reasoner, when required. The current Reasoner
implementation is based on Fact++
(owl.cs.manchester.ac.uk/tools/fact).
We also performed some user evaluations of
SemT++, which demonstrated that communication,
resource sharing, and shared resources retrieval with
SemT++ is significantly faster than without it, and
user satisfaction is higher. The details of this first
evaluation, together with the analysis and discussion
of the results, can be found in (Goy et al. 2014b).
Moreover, we evaluated the functionality of the User
Interface enabling the exploitation of multiple
criteria to perform object selection, and we found
that users actually appreciate it; see (Goy et al.
2014a).
4 ENHANCING SemT++ WITH
DOMAIN KNOWLEDGE:
DSemT++
Besides the knowledge modeling table resources as
information objects, represented in the Table
Ontology, DSemT++ tables have been equipped
with specific domain knowledge aimed at providing
a semantic characterization of the entities table
resources refer to, i.e. entities representing the
content of information objects.
The two properties defined in the Table
Ontology whose values refer to resource content are
hasTopic(x, y, t) and hasObjectOfDiscourse(x, y, t),
representing, respectively, the relation between an
information element (e.g., an email conversation)
and its main topic, and what a resource (e.g., a web
site) "talks about".
In the evaluation of the User Interface enabling
object selection based on multiple criteria,
mentioned above, many users claimed that the
meaning of some values of hasTopic and
hasObjectsOfDiscourse properties (typically added
ISE 2015 - Special Session on Information Sharing Environments to Foster Cross-Sectorial and Cross-Border Collaboration between Public
Authorities
374

by other table participants) can result unclear or
ambiguous, and they expressed the need of having
some explanations about the meaning of such
values.
The possibility of classifying an individual
representing a property value (e.g., Torino, as the
value of the hasTopic property of an article) in a
specific class (e.g., Municipality), and providing
other information about it (e.g., its location on a
map) could represent the "explanation" users were
asking for. It is worth mentioning that this support to
potentially ambiguous or unknown meanings of
property values is particularly important within a
collaborative environment such as DSemT++, where
a user can be unaware of the meaning of a property
value provided by another user.
To implement this functionality on DSemT++
tables, two semantic constituents are required: (a) a
Domain Ontology, modeling entities representing
topics and objects of discourse; (b) one or more
LOD dataset, containing data/information about the
chosen domain. These instruments, and knowledge
provided by LOD datasets in particular, besides
supporting the provision of "explanations" of the
meaning of topics and objects of discourse, also
offer the possibility of enriching table resources
themselves, by providing links to possibly related
resources; for example, if a document, lying on a
table concerning the organization of a music festival,
talks about the French composer Rameau, a link to
DBpedia could provide suggestions for adding
resources about baroque music on the table.
We thus improved the architecture of our system
by adding a Domain Knowledge Manager, which
manages the semantic knowledge concerning the
content of information objects (facts about the
individuals involved in the semantic representation
of resources content), which is stored in the domain
knowledge bases (Domain KBs); facts in such
knowledge bases are expressed according to the
corresponding Domain Ontologies; the set of
Domain Ontologies included in the system represent
the (static) semantic knowledge concerning the
domains table resources are about. Domain
Ontologies and Domain KBs are currently expressed
in OWL and the Domain Knowledge Manager
exploits the OWL API library to interact with them.
The Domain Knowledge Manager also invokes the
Reasoner, if required. Moreover, it handles the
connection with Linked Open Data (LOD) sets. To
this purpose, it exploits the Vocabulary Mappings
(mapping LOD datasets classes and properties onto
classes and properties belonging to system Domain
Ontologies), and the Instance Mappings (mapping
system and LOD datasets individuals). As we will
describe in Section 4.3, in the current prototype, the
Domain Knowledge Manager connects to the
GeoNames Search Web Service
(www.geonames.org/export).
4.1 Commonsense Geographic
Knowledge
DSemT++ tables and resources lying on them can
refer to a wide range of domains, so, in order to test
our approach, we had to choose a specific and well-
defined knowledge domain to be modeled in a
proof-of-concept prototype (see Section 4.3). We
considered commonsense geographic knowledge the
suitable domain to this purpose. In this perspective,
commonsense geographic knowledge is mainly
intended to be a testbed, since the whole framework
was designed to be reusable and to support data
models describing multiple knowledge domains,
possibly even on a single table.
However, besides being a testbed, commonsense
geographic knowledge has an intrinsic value. In fact,
together with time, space is one of the most
universal and cross-domain kinds of knowledge,
involved in a great number of different domains.
Commonsense geospatial knowledge comes in many
different ways into people's everyday life: we use
geographical concepts and relations when taking a
bus or a plane, when planning our holidays or when
arranging an appointment with someone. The
importance of geospatial knowledge in information
retrieval and in knowledge organization is also
claimed in the literature; see, for instance,
(Giunchiglia et al., 2012).
Further evidence of its centrality can be found in
the leading role geography has taken on in the
evolution of both the Web 2.0 and the Web of Data
(www.w3.org/2013/data) during the last ten years.
Services like Google Earth, Google Maps,
WikiMapia, and OpenStreetMap are enabling
geographically-based user-generated content.
Moreover, social networks like Foursquare, the
pervasive trend of geolocalization, and resource geo-
tagging increased the role of geography in our
everyday life. Simultaneously, the combination of
semantic technologies, the Web of Data and
Geographic Information resulted in the Semantic
Geospatial Web, a Semantic Web extension based
on several spatial ontologies, able to "increase the
relevance and quality of results in geographic
retrieval systems" (Ballatore et al. 2013, p. 95). In
such a process, the cross-domain nature of
geographic information acted as a "glue" in
A Semantic Framework to Enrich Collaborative Tables with Domain Knowledge
375

integrating and linking different datasets. The
connection role assumed by geographic information
in the Web of Data is further confirmed by a recent
report from the LOD workteam, where geography
appears as one of the nine thematic categories the
whole LOD cloud is divided in. In particular, this
latest crawl of the LOD cloud shows the role of hub,
together with DBpedia as general purpose dataset,
assumed by GeoNames during the last three years,
becoming de facto the reference geographic dataset
in the LOD scene (Schmachtenberg et al. 2014).
The Domain Knowledge Manager introduced
above, has thus been instantiated on commonsense
geographic knowledge, becoming the Geographic
Knowledge Manager. Before describing in detail
how it works, we will sketch a very simple usage
scenario (Section 4.2) in order to show how domain
knowledge (and in particular geographic knowledge)
can support table participants in building semantic
descriptions and in retrieving table resources on the
basis of their content.
4.2 Usage Scenario
The availability of geographic knowledge can help
DSemT++ users (at least) in two tasks: the creation
(and update) of semantic descriptions of table
objects, and the selection of criteria to retrieve them.
Consider the new object case (the update case is
similar): table participants can create a new object
(e.g., when they start writing a new document), or
they can add to the table an existing resource (e.g., a
bookmark pointing to a web page). In both cases, a
new semantic representation is built through the
following steps:
1. The Smart Object Analyzer automatically
determines the object formats (e.g., UTF-8,
HTML), its parts and their type (e.g., images
included in it); moreover, it proposes
candidate values for authors and languages the
information object is expressed in.
2. The Semantic Knowledge Manager, by
invoking the Reasoner, provides other
candidates for languages, for topics and for
objects of discourse (the set of candidates
suggested to users is the merge of the sets of
candidates proposed by the Smart Object
Analyzer and the set proposed by the
Semantic Knowledge Manager).
3. Users can confirm suggested values (i.e.,
candidate authors, languages, topics, and
objects of discourse), or they can select values
already used on the table for annotating other
objects, or they can introduce new ones.
Now, imagine that Roby participates in a table
concerning the activities of a small NGO for
environment safeguard, Save Our Earth, together
with some other volunteers. Roby has to write an
article for an online local newspaper, discussing the
situation of a local old farm building in
Champdepraz (a small municipality in Valle
d'Aosta). Roby creates a new table object (an HTML
document), writes some text in it, adds a picture of
the surrounding mountains and a hyperlink to a
resolution by the Municipality of Champdepraz
concerning a restoration project for the farm, aimed
at transforming it into a hotel. When Roby clicks the
"save&update" button, the creation of the object
semantic representation is triggered. The Smart
Object Analyzer (step 1) discovers that: the object
has a HTML representation, encoded in UTF-8; it
contains an image and a hyperlink; it may be written
in Italian; its author is probably Roby. The Reasoner
(step 2) infers the same candidate language, some
candidate topics (among them Champdepraz) and a
set of candidate objects of discourse (Champdepraz
farm building, restoration project, Ayasse river,
Mont Avic), mainly derived from topics and objects
of discourse of included objects (i.e., the image and
the resolution). Roby (step 3) confirms the language
(Italian), selects Champdepraz among the suggested
topics, and looks at the candidate objects of
discourse, in order to see if some of them could well
represent her article content. Roby knows the
restoration project by the Municipality for the old
farm building, close to the Ayasse river, but she is in
doubt about another suggested object of discourse,
i.e., Mont Avic: she knows there are a park and a
mountain with the same name; does the suggested
item refer to the mountain or to the park? Is it really
close to the Champdepraz farm building? Should she
mention it in the article and include it in the set of
objects of discourse representing the content of her
article?
Roby clicks on the suggested item (Mont Avic),
to get an explanation of it. The system displays a
pop-up window (see Figure 2) telling her that the
selected item refers to a 3.006 m. high mountain and
showing its position on a map. Moreover, a "more
information" link is available: when Roby clicks it,
she gets further data about Mont Avic. On the basis
of this information, Roby decides to add it as an
object of discourse of her article (in fact, although
currently it is not explicitly mentioned in the article,
the situation of a local old farm building in
Champdepraz definitely has a close relation with it).
Luca sits at the Save Our Earth table, looking for
pictures of Valle d'Aosta mountains, for a
ISE 2015 - Special Session on Information Sharing Environments to Foster Cross-Sectorial and Cross-Border Collaboration between Public
Authorities
376

Figure 2: "Explanation" of an object of discourse in
DSemT++.
photographic reportage he is going to create. He
selects topics as the first criterion for object
selection; the table presents him a list of table topics,
among which Mont Avic; Luca wonders if it is
intended as the mountain or the park: he clicks on
the item and gets the "meaning" of the topic Mont
Avic, i.e., all the information available for it (see
Figure 2); such information enables him to discover
that it refers to the mountain, and thus he selects it,
so getting a very nice picture of Mont Avic, useful
for his reportage.
4.3 The Role of the Geographic
Knowledge Manager
As we mentioned, in order to provide tables with
geographic knowledge, we instantiated the Domain
Knowledge Manager module onto the geographic
domain, creating the Geographic Knowledge
Manager. The instruments we need to implement the
Geographic Knowledge Manager functionality are a
Geographic Ontology and a suitable geographic
LOD dataset: in the following we will describe
them.
Geographic Ontology
The semantic model of the geographic domain is
provided in the Geographic Ontology. This
component represents the system view of the
geographic domain and its role consists in providing
a vocabulary to describe the content of table
resources (as far as the geographic aspects are
concerned). In other words, the Geographic
Ontology provides the conceptual view enabling the
system to "interpret", and thus integrate, data
belonging to potentially heterogeneous sources.
The Geographic KB contains all the "facts", i.e.
semantic assertions, about geographic instances:
each new geographic instance in DSemT++ (e.g., the
instance representing Mont Avic) is classified with
respect to the Geographic Ontology (e.g., as an
instance of the Mountain class).
The Geographic Ontology is a lightweight, task-
and application-oriented ontology, containing about
240 classes and a number of properties, mainly
reflecting the properties used by GeoNames to
describe features (such as latitude, longitude,
population, altitude, etc.).
Two classes represent the top layer of the
taxonomy:
 GeoSocialEntity includes all those geospatial
entities whose existence is due to people's
activities; it encompasses concrete entities, like
infrastructures and human settlements, as well
as concepts usually used to partition the
geographic space, and administrative or
political institutions.
 GeoPhysicalEntity includes all natural or
geophysical entities like rivers, mountains,
deserts, gulfs, valleys, and so on.
Although the Geographic Ontology partially
reflects the GeoNames ontology (see below), it is an
independent semantic model. DSemT++, in fact, is
not committed to any specific external geographic
dataset and thus the Geographic Ontology, by
providing the system with a conceptual view over
the geographic domain, enables the integration
within the system of geographic data coming from
different datasets and possibly originally
characterized by means of different ontologies.
Thus, DSemT++ Geographic Ontology, along with
the suited mappings (see Vocabulary Mappings
section below), represents a unifying view over
heterogeneous geographic semantic models,
exploited in the LOD cloud.
The GeoNames Dataset
First released in 2006, GeoNames is an open
geospatial gazetteer gathering different official data
sources (mainly from governmental organizations,
institutes of geography and statistics) and combining
them with users' contribution. The GeoNames
A Semantic Framework to Enrich Collaborative Tables with Domain Knowledge
377

database contains over 10 millions of toponyms and
9 millions of features, 2.8 millions of which are
populated places. The features are classified
according to an OWL taxonomy, the so-called
GeoNames ontology, made up of 9 high-level
classes, called Feature Classes, and 650 subclasses,
called Feature Codes. Each GeoNames feature is
uniquely identified by an URI and the whole
gazetteer is available both in RDF and as database
dump. Moreover, GeoNames makes available
RESTful Web Services (www.geonames.org/export/
ws-overview.html) enabling different types of
queries; for example, besides a general purpose
search service, search for closest toponyms, altitude
of a geographic point, cities and toponyms within a
user specified bounding box, postal codes,
earthquakes, timezones. All services can be invoked
via HTTP GET requests; the most part of the results
are returned by GeoNames as an XML or JSON
object, while for the search service it is also possible
to obtain the results as RDF.
In DSemT++, we employed the searchJSON
service, i.e., the general purpose search service
returning a list of results in JSON format.
Vocabulary Mappings
In order to be exploited in DSemT++, the
Geographic Ontology and GeoNames need to be
"linked", so that the entities of the latter could be
classified into classes of the former. We thus defined
a mapping between the entities of the GeoNames
ontology (o
GN
) and the entities of our Geographic
Ontology (o
GO
), relying on the following two
relations:
 o
GN
= o
GO
 o
GN
< o
GO
Two cases are possible: o
GN
is a feature code
represented in the GeoNames ontology and o
GO
is a
class of the DSemT++ Geographic Ontology, or o
GN
and o
GO
are both properties, the former belonging to
the GeoNames ontology and the latter to the
DSemT++ Geographic Ontology. Moreover, =
expresses conceptual equivalence, and < expresses
the fact that the right-hand side concept subsumes
the left-hand side one. For example, Figure 3 shows
the RDF/XML serialization of the axiom which
states the subclass relationship between the class
representing all individuals having H.STMH as
Feature Code value in GeoNames ontology and the
class WaterSpring in the Geographic Ontology.
These axioms enable us to achieve the goal of
making the two ontologies intelligible to one
another, and thus being able to import knowledge
from the GeoNames dataset into our system.
DSemT++ Vocabulary Mappings mention 192
classes from the Geographic Ontology and 233
Feature Codes from the GeoNames ontology,
establishing 186 equivalence axioms and 31
subsumption axioms.
<owl:Restriction>
<rdfs:subClassOf
rdf:resource="http://www.di.unito.it/onto
logies/SemTppOntologies/
SemTppGeographicOntology#WaterSpring"/>
<owl:onProperty
rdf:resource="http://www.geonames.org/ont
ology#featureCode"/>
<owl:hasValue
rdf:resource="http://www.geonames.org/ont
ology#H.STMH"/>
</owl:Restriction>
Figure 3: The axiom stating the subclass relationship
between a class of the GeoNames ontology and a class of
the DSemT++ Geographic Ontology.
Geographic Knowledge Manager
The role of the Geographic Knowledge Manager is
twofold:
 It interacts with GeoNames to retrieve
information about geographic entities, i.e.
about topics and objects of discourse of table
resources. The GeoManager submodule,
shown in Figure 4, is in charge of this activity.
 It interacts with the Geographic Ontology and
the Geographic KB and invokes the Reasoner
to classify the GeoNames entities obtained at
the previous step under the Geographic
Ontology schema. In order to achieve this goal
it exploits the Vocabulary Mappings (described
above). The OntoMgmService submodule,
shown in Figure 4, is responsible of this
activity.
The GeoManager and the OntoMgmService have
been implemented, in the proof-of-concept
prototype, using different technologies: the
asynchronous web framework Node.js for the
former, and Java Servlets, exploiting the OWL API
library, for the latter. This choice has been mainly
suggested by the interactions these modules have
with datasets and knowledge bases, i.e. the external
dataset GeoNames for the GeoManager and the
OWL local ontology and KB for the
OntoMgmService. Given such heterogeneity, we
ISE 2015 - Special Session on Information Sharing Environments to Foster Cross-Sectorial and Cross-Border Collaboration between Public
Authorities
378

designed the OntoMgmService as a RESTful
service, accessible through HTTP requests and
exchanging data in JSON format. In particular, the
OntoMgmService is identified by a URL; the
GeoManager invokes it (see step 4(b) below) by
sending a POST HTTP request which contains a
JSON object (whose main element is the system IRI
identifying the geographic instance representing the
topic/object of discourse in focus). The
OntoMgmService, written in Java, invokes the
Reasoner (currently Fact++) in order to classify the
instance in the suited class of the Geographic
Ontology.
Figure 4: The Geographic Knowledge Manager
architecture.
Moreover, information retrieved from GeoNames is
stored in a local database (Local Geo DB),
implemented in MongoDB (www.mongodb.org).
To provide a better understanding of the
Geographic Knowledge Manager functionality, we
describe its behavior in a typical use case:
1. The GeoManager receives from the TO
Manager a string corresponding to a new topic
or object of discourse (e.g., "Mont Avic"),
together with the IRI referring to the instance
created by the system for that topic/object of
discourse (e.g., http://www.di.unito.it/semtpp/
resources/mont_avic). The string is used as a
keyword to query the GeoNames dataset,
through the searchJSON service.
2. GeoNames returns a JSON object containing a
list of entities, along with their descriptions.
3. The GeoManager sends these results back to
the TO Manager, which passes them to the
User Interaction Manager, thus enabling the
user to select the proper entity, if any.
4. The system IRI of the instance representing
the new topic/object of discourse, together
with the GeoNames ID, are sent to the
GeoManager, which: (a) uses the GeoNames
ID to check if GeoNames data about the entity
are already present in the Local Geo DB, and
add them if not; (b) uses the system IRI to
invoke the OntoMgmService, in order to have
the instance classified with respect to the
Geographic Ontology (e.g., classifying Mont
Avic as an instance of Mountain).
In this way, the external semantic knowledge
available in LOD sets (GeoNames in our prototype)
is brought into the system, linked to the semantic
description of table resources (as depicted in Figure
5), and available to table users: when a table user
clicks on that topic/object of discourse, the result of
the instance classification, together with other
relevant GeoNames data (e.g., localization on a
map), are displayed (see Figure 2, where the
information about Mont Avic is shown).
Figure 5: Adding geographic knowledge to semantic
descriptions of table objects: an example.
As we shown in the usage scenario (Section 4.2),
this knowledge provides table users with an
"explanation" of the meaning of the topics/objects of
discourse, which can be useful at least in two cases:
when annotating table resources with semantic
properties representing their content (i.e., topic and
objects of discourse), and when selecting table
objects on the basis of their content. Moreover,
knowledge retrieved from LOD datasets can be
exploited to enrich table resources by providing
links to possibly related new contents (e.g., a link to
the Gran Paradiso massif in case of a resource
talking about Mont Avic).
4.4 Preliminary Evaluation
Since the enhancement of SemT++ with domain
knowledge started from a need that users pointed out
while evaluating our first prototype (Goy et al.,
2014a), following a user-centered design approach,
we contacted again the same 20 participants of the
test which evaluated the use of multiple criteria to
select table objects, and we asked them to perform
the same task, paying attention to the fact that now
an "explanation" is available for some topics and
objects of discourse (i.e., for those related to
A Semantic Framework to Enrich Collaborative Tables with Domain Knowledge
379

geographic features). Since participants represent
potential (D)SemT++ users, we ensured that they
were familiar with the system already in the first
evaluation.
We asked participants to rate this new
functionality, on a 1 to 5 scale. We obtained an
average of 4.45, indicating that the new feature was
appreciated by users (the low standard deviation tells
us that users tend to agree on it). In the free
comments section of the brief questionnaire, some
users told us that the functionality would be more
interesting if not only geographic issues were
supported. On the basis of this − quite obvious −
observation, we are going to extend the prototype in
order to connect other LOD datasets.
5 CONCLUSIONS
In this paper we presented DSemT++, an
environment supporting users in the collaborative
management of heterogeneous resources, enhanced
with domain knowledge partially retrieved from
LOD datasets.
We did not explicitly faced here all the issues
concerning collaboration, both regarding resource
handling and regarding collaborative metadata
management. These aspects are discussed in (Goy et
al. 2015). Moreover, also some issues concerning
the management of semantic knowledge in
DSemT++ deserve further study. For example, we
are investigating how information and links
retrieved from LOD datasets can be used to provide
users with suggestions about content items related to
the resource in focus, taking into account also the
context represented by the activity the table is
devoted to. Moreover, the connection of new
datasets to DSemT++ currently requires, in many
cases, the manual definition of the local Domain
Ontology and the Vocabulary Mappings. It would be
interesting to investigate the possibility of a semi-
automatic support for the integration of ontologies
underlying LOD datasets; see, for instance, (Zhao
and Ichise, 2014). Furthermore, we are planning a
new evaluation of DSemT++ with users, in order to
assess the usefulness of domain knowledge within
the system.
Finally, we would like to investigate the
applicability of the proposed approach to other
contexts, in particular to the management of archival
resources. Semantic knowledge represented by
ontologies and data from the LOD cloud, in fact,
could represent precious instruments to enhance the
access and management of such resources.
REFERENCES
Abel, F., Henze, N., Krause, D., Kriesell, M., 2010.
Semantic enhancement of social tagging systems. In
V. Devedžić, D. Gašević (Eds.), Web 2.0 & Semantic
Web, Heidelberg: Springer, 25–56.
Ardissono, L., Bosio, G., Goy, A., Petrone, G., 2010.
Context-aware notification management in an
integrated collaborative environment. In A. Dattolo, C.
Tasso, R. Farzan, S. Kleanthous, D. Bueno Vallejo, J.
Vassileva (Eds.), Int. Workshop on Adaptation and
Personalization for Web 2.0, vol. 485, CEUR, 21–30.
Ballatore, A., Wilson, D. C., Bertolotto, M., 2013. A
survey of volunteered open geo-knowledge bases in
the semantic web. In G. Pasi, G. Bordogna, L.C. Jain
(Eds.), Quality issues in the management of web
information, Heidelberg: Springer, 93–120.
Bardram, J. E., 2007. From desktop task management to
ubiquitous activity-based computing. In V. Kaptelinin,
M. Czerwinski (Eds.), Beyond the Desktop Metaphor,
Cambridge, MA: MIT Press, 223–260.
Barreau, D.K., Nardi, B., 1995. Finding and reminding:
File organization from the desktop. ACM SIGCHI
Bulletin, 27(3), 39–43.
Borgo, S., Masolo, C., 2009. Foundational choices in
DOLCE. In S. Staab, R. Studer, R. (Eds.), Handbook
on ontologies, second edition, Heidelberg: Springer,
361–381.
Breslin, J. G., Passant, A., Decker, S., 2009. The social
semantic web. Heidelberg: Springer.
Drăgan, L., Delbru, R., Groza, T., Handschuh, S., Decker,
S., 2009. Linking semantic desktop data to the Web of
Data. In L. Aroyo, C. Welty, H. Alani, J. Taylor, A.
Bernstein, L. Kagal, N. Noy, E. Blomqvist (Eds.), The
Semantic Web – ISWC 2011, LNCS 7032, Heidelberg:
Springer, 33–48.
Evangelou, C., Karacapilidis, N., 2005. On the interaction
between humans and Knowledge Management
Systems: A framework of knowledge sharing
catalysts. Knowledge Management Research and
Practice, 3(4), 253–261.
Gangemi, A., Borgo, S., Catenacci, C., Lehmann, J., 2005.
Task taxonomies for knowledge content. Metokis
Deliverable D07.
Giunchiglia, F., Dutta, B.,·Maltese, V., Feroz, F., 2012. A
facet-based methodology for the construction of a
large-scale geospatial ontology, J. of Data Semantics,
1(1), 57–73.
Goy, A., Magro, D., Petrone, G., Picardi, C., Segnan, M.,
2015. Ontology-driven collaborative annotation in
shared workspaces. Future Generation Computer
Systems, in press.
Goy, A., Magro, D., Petrone, G., Segnan, M., 2014(a).
Semantic representation of information objects for
digital resources management. Intelligenza Artificiale,
8(2), 145–161.
Goy, A., Petrone, G., Segnan, M., 2014(b). A cloud-based
environment for collaborative resources management.
Int. J. of Cloud Applications and Computing, 4(4), 7–
31.
ISE 2015 - Special Session on Information Sharing Environments to Foster Cross-Sectorial and Cross-Border Collaboration between Public
Authorities
380

Hu, W., Jia, C., Wan, L., He, L., Zhou, L., Qu, Y., 2014.
CAMO: Integration of Linked Open Data for
multimedia metadata enrichment. In P. Mika, T.
Tudorache, A. Bernstein, C. Welty, C. Knoblock, D.
Vrandĕcíc, P. Groth, N. Noy, K. Janowicz, C. Goble
(Eds.), The SemanticWeb – ISWC 2014, LNCS 8796,
Heidelberg: Springer, 1–16.
Kaptelinin, V., Czerwinski, M., 2007. Beyond the desktop
metaphor. Cambridge, MA: MIT Press.
Karger, D.R., 2007. Haystack: Per-user information
environments based on semistructured data. In V.
Kaptelinin, M. Czerwinski (Eds.), Beyond the Desktop
Metaphor, Cambridge, MA: MIT Press, 49–100.
Kim, H., Breslin, J.G., Decker, S., Choi, J., Kim, H., 2009.
Personal Knowledge Management for knowledge
workers using social semantic technologies. Int. J. of
Intelligent Information and Database Systems, 3(1),
28–43.
Magro, D., Goy, A., 2012. A core reference ontology for
the customer relationship domain. Applied Ontology,
7(1), 1–48.
Meimaris, M., Alexiou, G., Papastefanatos, G., 2014.
LinkZoo: A linked data platform for collaborative
management of heterogeneous resources. In Presutti
V., d'Amato C., Gandon F., d'Aquin M., Staab S.,
Tordai A. (Eds.), The Semantic Web: Trends and
Challenges, LNCS 8465, Heidelberg: Springer, 407–
412.
Passant, A. Laublet, P., 2008. Meaning Of A Tag: A
collaborative approach to bridge the gap between
tagging and Linked Data. In C. Bizer, T. Heath, K.
Idehen, T. Berners-Lee (Eds.), Linked Data on the
Web (LDOW 2008), vol. 369. CEUR.
Sauermann, L., Bernardi, A., Dengel, A., 2005. Overview
and outlook on the semantic desktop. In S. Decker, J.
Park, D. Quan, L. Sauermann (Eds.), Semantic
Desktop Workshop, vol. 175. CEUR.
Schandl, B., Haslhofer, B., Bürger, T., Langegger, A.,
Halb, W., 2012. Linked Data and multimedia: The
state of affairs. Multimedia Tools and Applications,
59(2), 523–556.
Schmachtenberg, M., Bizer, C., Paulheim, H., 2014)
Adoption of the Linked Data best practices in different
topical domains. In P. Mika, T. Tudorache, A.
Bernstein, C. Welty, C. Knoblock, D. Vrandĕcíc, P.
Groth, N. Noy, K. Janowicz, C. Goble (Eds.), The
SemanticWeb – ISWC 2014, LNCS 8796, Heidelberg:
Springer, 245–260.
Voida, S., Mynatt, E. D., Edwards, W. K., 2008. Re-
framing the desktop interface around the activities of
knowledge work. In Proc. of UIST'08, New York, NY:
ACM Press, 211–220.
Yamada, I, Ito, T., Usami, S., Takagi, S., Toyoda, T.,
Takeda, H., Takefuji, Y., 2014. Linkify: Enhanced
reading experience by augmenting text using Linked
Open Data. ISWC 2014 Semantic Web Challenge:
challenge.semanticweb.org.
Zhao, L., Ichise, R., 2014. Ontology integration for Linked
Data, J. of Data Semantics
, 3(4), 237–254.
A Semantic Framework to Enrich Collaborative Tables with Domain Knowledge
381