Unlocking Serendipitous Learning by Means of Social Semantic Web

Matteo Gaeta

, Vincenzo Loia

, Giuseppina Rita Mangione

, Sergio Miranda

and Francesco Orciuoli

DIEM, University of Salerno, Via Giovanni Paolo II, 132, Fisciano (SA), Italy

DI, University of Salerno, Via Giovanni Paolo II, 132, Fisciano (SA), Italy

Keywords:

Serendipitous Learning, Social Semantic Web, SKOS, SIOC.

Abstract:

Serendipitous Learning is the learning process occurring when hidden connections or analogies are unexpect-

edly discovered, mostly during searching processes (for instance on the Web) which are typical for informal

learning activities, especially accomplished in the workplace context. Moreover, serendipitous processes have

high probability to occur in the contexts where learners have high autonomy, more chances to intervene in

different activities and to interact with resources and people. This paper proposes an approach, based on the

Social Semantic Web, to sustain and foster Serendipitous Learning. The proposed approach considers two con-

nected ontology layers to model knowledge by using several Semantic Web vocabularies like SIOC, Dublin

Core, SKOS, and so on. The SKOS role is particularly relevant because it allows connections among hetero-

geneous resources, also across multiple communities. The proposed approach models the above-mentioned

connections at the conceptual level in order to facilitate learners to discover relevant links, concepts and to

follow unexpected useful paths.

1 INTRODUCTION AND

RELATED WORKS

Technologies and, in particular, ICTs (Information

and Communication Technologies) inﬂuence the ed-

ucational practices as well as the redeﬁnition of the

concepts of learning and teaching. This phenomenon

asks educators to review their practices in order to

face new learning/teaching situations in which learn-

ers can access alternative learning resources in ad-

dition to those deﬁned in structured curricula. The

availability of new learning resources is mainly due

to the Web, intended as a boundless ﬁeld of informa-

tion. The increasing number of learning resources (on

the Web) and the decreasing of teacher presence re-

quire higher autonomy and regulation to learners in

order to master both formal and informal learning en-

vironments. Autonomous and self-directed interac-

tions with available content foster ”[...] unexplored

and unplanned discoveries and fortunate incidents in

the process of exploring something else [...]” (Kop,

2012). In this context, the term serendipitous learning

was coined to point out the learning processes related

to the ”[...] unexpected realization of hidden, seem-

ingly unrelated connections or analogies [...]” (Kop,

2012).

This paper describes how Social Semantic Web, if

opportunely deployed, is able to effectively sustain

serendipitous learning. Social Semantic Web (Bres-

lin et al., 2009) is deﬁned as the synergistic applica-

tion of Semantic Web and Social Web that tries to get

and combine the advantages of both. The work pro-

vides a scalable approach to manage knowledge and

contents created in a speciﬁc environment. It mainly

focuses on both the formal description of relevant re-

sources (e.g. Web resources) in the environment and

the modelling of conceptual connections among the

above-mentioned resources. The approach proposes

a knowledge representation architecture based on two

integrated layers of ontologies in order to foster the

emergence of relevant connections among resources

and the agile provisioning of these connections to the

learners, helping them to play a serendipitous learn-

ing experience. The approach could be exploited

also in smaller contexts like, for instance, Small and

Medium-sized Enteprises (SMEs), where the concept

of Web resource is replaced by the concept of digital

resource. Other works in literature deal with the adop-

tion of the Social Semantic Web as a learning plat-

form. In particular, the authors of (Jovanovic et al., ),

(Torniai et al., 2008) and (Jeremic et al., 2013) were

among the ﬁrst ones to describe the e-learning scenar-

285

Gaeta M., Loia V., Rita Mangione G., Miranda S. and Orciuoli F..

Unlocking Serendipitous Learning by Means of Social Semantic Web.

DOI: 10.5220/0004845102850292

In Proceedings of the 6th International Conference on Computer Supported Education (CSEDU-2014), pages 285-292

ISBN: 978-989-758-020-8

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

ios supported and enhanced by the Social Semantic

Web technologies and methodologies and to discuss

the main issues concerning the realization of these

scenarios:

• development and maintenance of domain ontolo-

gies;

• exploitation of user-generated contents as learn-

ing resources;

• provisioning of new forms of interaction;

• support for interoperability;

• support for adaptation and personalization of e-

learning experiences;

• ubiquitous access to learning resources.

Moreover, the authors of (Brooks et al., ) provide

three main lessons learned by analysing e-learning

applications and tools developed by using Semantic

Web and Web 2.0 technologies:

• testing the deﬁned approaches in real world sce-

narios in order to effectively evaluate systems and

validate models and methodologies;

• tracking large amount of data related to learners’

interactions in order to extract pedagogical pat-

terns;

• extracting implicit knowledge from contents in

order to automatically construct metadata to im-

prove search of learning material.

With respect to the reviewed literature, this paper

mostly focuses on the anatomy of the ontology struc-

tures (two integrated layers of ontologies have been

introduced and modelled) and on the methods for

constructing, maintaining and evolving the aforemen-

tioned ontologies. More details concerning the dis-

covery of links and related resources have been pro-

vided by proposing several approaches. If the Seman-

tic Web technologies provide a great solution for in-

teroperability and integration, discovery is considered

one of the main lever to enable serendipity. This paper

takes care of principles and suggestions of the exist-

ing works and advances by trying to solve some of the

main issues related to the adoption of Social Semantic

Web in e-learning scenarios. Lastly, it is important to

underline that the proposed approach aims at working

in contexts (e.g. Knowledge-Intensive Organizations)

where the ICTs are already exploited to manage and

track (also partially) the work activities.

2 SERENDIPITOUS LEARNING

The idea underlying serendipitous learning is based

on several pedagogical approaches. Among the oth-

ers, it is possible to recognize discovery learning, ex-

ploratory learning, experiential learning, construc-

tivist learning and connectivism (Kop, 2012). Nu-

merous works (Heinstr

om, 2007) address serendipi-

tous learning by taking care mostly of the concept of

surprise and on the accidental nature of information

discovery. On the other hand, some authors (Andr

et al., 2009) focus on the importance of prior knowl-

edge and sagacity in the serendipitous learning pro-

cesses. Other scientiﬁc results afﬁrm that the hidden

connection discovery could not occur instantly but re-

quire an incubation period to the learners (McCay-

Peet and Toms, 2010) (Lu, 2012). These authors as-

sert also that the conditions able to sustain serendipi-

tous connection discovery are those of active learning

and social learning in knowledge building and discov-

ery environments (Scardamalia and Bereiter, 2006)

(McCay-Peet and Toms, 2010). Furthermore, the ef-

fects of serendipity are analysed by the authors in

(Gritton, 2007). They assert that there is not sufﬁ-

cient evidence to afﬁrm that serendipitous learning

is a consequence of intuitive sagacity of learners but

they state that there is no doubt that serendipitous

browsing can reveal hidden connections among con-

cepts and stimulate thinking and, consequently, learn-

ing. Now, the question is: may the new Web technolo-

gies increase and improve serendipitous learning?.

Some ideas are proposed by authors of (Ihanainen

and Moravec, 2011) and (Boyd, 2010). They en-

hance search strategies (e.g. the same adopted by Web

search engines) by moving the control from search en-

gines to learners and by fostering randomness in the

information stream. Authors of (Boyd, 2010) afﬁrm

also that people should ﬁnd methods for integrating

Web searching into their thinking and reﬂection pro-

cesses. Technologies have to take care of this integra-

tion and the personal context of the learner by consid-

ering an unﬁltered but manageable store of resources.

3 A SOCIAL SEMANTIC WEB

APPROACH TO IMPROVE

SERENDIPITOUS LEARNING

In this Section, an approach that fosters serendipitous

learning is proposed. The approach is based on the

main Social Semantic Web principles in order to pro-

vide Web-based environments to interact (searching,

tagging, rating, recommending, etc.) with digital re-

sources.

CSEDU2014-6thInternationalConferenceonComputerSupportedEducation

286

3.1 The Social Semantic Web

The W3C

provides the following brief deﬁnition for

Semantic Web: [..] The Semantic Web provides a

common framework that allows data to be shared and

reused across application, enterprise, and community

boundaries [..]. The Semantic Web vision concerns

the provisioning of metadata associated with Web re-

sources to assign machine-interpretable meaning to

them. In order to make this metadata really sharable

and understandable, it requires vocabularies and on-

tologies expressing semantics for terms and values

used in the above-mentioned metadata. Being these

metadata set represented in uniform way, better link-

ing and decentralization of resources are supported.

Among the other issues, the Semantic Web infrastruc-

ture provides concrete answers to the requests for en-

tity identity and explicit relationships. More in de-

tails, with Semantic Web, the entities on the Web,

previously embedded and hidden in HTML pages, be-

come uniquely identiﬁable and, thus, understandable

and manageable by machines (computers). Moreover,

representing entities is not sufﬁcient, thus identifying

relationships among entities is also needed (Breslin

et al., 2009). In other words, Semantic Web allows the

development of software agents that aim at improv-

ing search and navigation of Web resources, making

new user experiences available. From the technologi-

cal viewpoint, the core of the Semantic Web architec-

ture

is represented by: i) RDF (Resource Descrip-

tion Framework), i.e., a data model natively enabling

distribution of data, ii) RDFS (Resource Description

Framework Schema) and OWL/OWL2 (Web Ontol-

ogy Language), i.e., languages for formal description

of data and its semantics. These two main layers,

supported by URIs and XML, allow querying (by us-

ing, for instance, SPARQL) and inference (by using

ontology-based inference enabled by OWL reasoners

or rule-based inference enabled by speciﬁc rule en-

gines). On the top of the core layers, a great num-

ber of vocabularies and ontologies, covering different

domains (some more speciﬁc, others more general),

have been developed and deployed. FOAF (Friend of

a Friend)

for representing knowledge on people and

social relationships, SIOC (Semantically-Interlinked

Online Communities)

for representing knowledge on

activities of on-line communities, Dublin Core for de-

scribing metadata for (digital) resources, SKOS (Sim-

ple Knowledge Organization System)

for modelling

http://www.w3.org/2001/sw/

http://www.w3.org/standards/semanticweb/

http://www.foaf-project.org

http://sioc-project.org

http://www.w3.org/2004/02/skos/

thesauri, concept maps and controlled vocabularies,

SCOT (Social Semantic Cloud of Tags)

& MOAT

(Meaning of a Tag Ontology) (Passant and Laublet,

2008) for modelling all the entities related to the tag-

ging process, and so on.

In this context, the concept of Social Semantic

Web emerges. Social Semantic Web represents the

ecosystem in which social interactions on the Web

applications lead to the creation of explicit and se-

mantically rich knowledge representations. The So-

cial Semantic Web combines technologies, strategies

and methodologies from Semantic Web, social soft-

ware and the Web 2.0 (Weller, 2010). In other words,

the Semantic Web provides new and better searching

(for people, content, tag, etc.) scenarios for Social

Web applications and, vice-versa, users’ interactions

with Social Web applications extend, update and pop-

ulate the Semantic Web structures.

SIOC (Boj

ars et al., 2008) is one of the most im-

portant and known ontologies for the Social Semantic

Web. It aims at interlinking on-line user-generated

content from applications such as blogs, instant mes-

saging tools, wikis and other Social Web sites, by pro-

viding an ontology to model structures (e.g. posts,

threads, etc.) and activities (e.g. tagging, reply-

ing, etc.) in online communities. Typically, SIOC

is used in combination with the FOAF vocabulary

for describing people and their friends and the SKOS

model for organising knowledge. SIOC lets develop-

ers link discussion posts and content items to other

related discussions, items, people and topics. Authors

of (Boj

ars et al., 2008) introduce also the concept of

food chain (see Fig. 1).

Figure 1: SIOC food chain.

The idea is that semi-structured data is gath-

ered and transformed in SIOC data from different

sources. Subsequently, SIOC data coming from dif-

ferent sources is aggregated (by using the native ca-

pabilities of RDF data model) and indexed (e.g. with

SKOS taxonomies). At the end of the chain, users can

search and navigate integrated SIOC structures also

by surﬁng the deﬁned indexes.

http://scot-project.net/scot/spec/scot.html

UnlockingSerendipitousLearningbyMeansofSocialSemanticWeb

287

3.2 Generalizing the SIOC Approach

It is possible to generalize the SIOC (Breslin et al.,

2006) approach by considering different types of data

source and not only Social Web applications. The

idea is to provide additional ontologies, to integrate

with SIOC, to cover different domains. For instance

it is possible to use Dublin Core to describe docu-

ments in Document Management Systems (DMSs),

IEEE LOM (a binding in RDF

) to describe learning

objects in Learning Object Repositories (LORs) and

so on. Fig. 2 shows a generalization of the SIOC food

chain and proposes SKOS as a mechanism to integrate

and index different types of individuals.

Figure 2: Generalizing the SIOC food chain.

Fig. 2 clearly shows two ontology layers, the up-

per layer (in Fig. 2), realized by means of SKOS,

provides classiﬁcation ontologies, i.e., lightweight se-

mantic structures used to annotate and classify in-

dividuals belonging to the ontologies at the lower

layer. The lower layer provides descriptive ontolo-

gies which model key concepts of the domain of in-

terest. For instance, in a Knowledge-Intensive Orga-

nization (KIO), the key concepts like Project, Task,

Activity, Document, UGC (User-Generated Content)

should be considered and modelled by using the on-

tologies at the lower level (Mangione et al., 2012)

(Gaeta et al., 2013). These ontologies are, relatively,

static in the sense that they do not rapidly evolve and

have to be built and managed by human experts fol-

lowing methodologies like NeON (Su

arez-Figueroa,

2010) also by adopting (as suggested in this work)

existing vocabularies. Ontologies at the upper layer

are dynamic, in the sense that they evolve as soon

as new activities (dealing with new topics) start in

the KIO. These ontologies model and structure, for

http://ltsc.ieee.org/wg12/

instance, the topics of the projects in the KIO. The

classiﬁcation ontologies can be constructed by means

of automatic, semi-automatic or non-automatic pro-

cesses. More in details, the automatic process can be

realized by means of Fuzzy Formal Concept Analysis

(FFCA) (Maio et al., 2011) that is used to extract con-

cepts from a set of full-text documents and construct

a Fuzzy Concept Lattice, i.e., a structure that repre-

sents the conceptualization of the document set taken

as input. The same authors deﬁne a methodology to

build SKOS ontologies starting from a Fuzzy Con-

cept Lattice. If the considered documents are pack-

aged as structured learning contents it is possible to

adopt approaches like, for instance, those provided

in (Capuano et al., 2009) and (Gaeta et al., 2011),

where the content structure is exploited to construct

lightweight ontologies from the source learning mate-

rial. Deﬁnitely, the combination of the two ontology

layers allows interoperability, integration (in particu-

lar descriptive ontologies) and serendipity (in partic-

ular classiﬁcation ontologies) as we will show in the

next Sections.

3.3 Using SKOS to Link Heterogeneous

Contents

In the Semantic Web, SKOS provides a vocabulary

to deﬁne thesauri, taxonomies, controlled vocabular-

ies, concept maps and so on. SKOS provides the

skos:Concept class that can be instantiated in or-

der to deﬁne individual concepts (or subjects, top-

ics, tags, etc.) which can be hierarchically related by

means of skos:narrower and skos:broader (Baker

et al., 2013). Other relations among individual con-

cepts can be expressed by means of skos:related

(for weak semantic relations) or by deﬁning sub-

properties of the above mentioned properties, includ-

ing skos:semanticRelation. Moreover, two con-

cepts belonging to two different SKOS taxonomies

can be related by using skos:relatedMatch,

narrowMatch and broadMatch.

According to the need of indexing and integrating

individuals coming from different classes (of different

vocabularies), it is important to introduce two prop-

erties: sl:isSubjectOf and sl:subject. The ﬁrst

one is the inverse of the second one that is deﬁned as

a subproperty of dct:subject

(coming from Dublin

Core vocabulary).

Moreover, sl:subject has range skos:Concept

whilst sl:isSubjectOf has domain skos:Concept.

More in details, as shown in Fig. 3, by means of the

sl:isSubjectOf property it is possible to assert that

http://dublincore.org/documents/2012/06/14/

dcmi-terms/?v=terms#terms-subject

CSEDU2014-6thInternationalConferenceonComputerSupportedEducation

288

Figure 3: SKOS to link heterogeneous contents.

both the document d and the (forum) post p deal with

the topic Dublin Core. The post p also deals with

the topic Semantic Web. The concept Dublin Core

is related to the concept DMS of a different concept

schema, as well as the concept Enterprise Apps is

related to the same concept DMS.

Deﬁnitely, as we will see also in the next Section,

the classiﬁcation ontologies represent the core struc-

ture enabling the serendipitous discovery.

3.4 Serendipitous Learning in the Social

Semantic Web

According to the example provided in Fig. 3, as-

sume that user A is interested in the concept DMS be-

cause she has to introduce a Document Management

System in the SME (Small and Medium-sized Enter-

prise) she works for. While addressing the DMS con-

cept, Dublin Core concept is signalled to A because

it is related to DMS. Once the attention of A is focused

on Dublin Core, she learns that Dublin Core is re-

lated to DMS (we assume that a standard viewer, for

a SKOS ontology, shows and emphasizes at least all

the concepts directly linked to the concept the user is

currently addressing) and she can see two resources

(coming from two different systems) dealing with the

aforementioned concept. By reading the text docu-

ment d, A goes deep into her knowledge on the con-

nection among Dublin Core and DMS. A learns that

some existing Document Management Systems rep-

resent document metadata by means of Dublin Core

schema.

Moreover, A, by reading the forum post p under-

stands that the use Dublin Core follows the principle

of the Semantic Web, a concept that is unknown for

A until this moment. A acquires knowledge about the

Semantic Web concept by enjoying the learning ob-

ject lo.

This sample ﬂow emphasizes the main principles

of Serendipitous Learning:

• Fostering autonomous interaction of users-

learners with content space;

• Valorizing the users-learners’ prior knowledge (in

the out-of-the-box thinking) by means of SKOS

concept schemas which recall the human mind

model;

• Stimulating surprise (by showing hidden connec-

tions) during exploratory processes;

• Providing active and social learning space aiming

at knowledge building processes.

The relations among concepts (skos:Concept)

and Web resources (i.e. the Aggregating and Index-

ing process of Fig. 2), in the Social Semantic Web

platform could be generated with non-automatic, au-

tomatic and semi-automatic mechanisms. In the ﬁrst

case, it is possible to exploit the tagging features of

Social Web applications and integrate tagging ontolo-

gies like, for instance, SCOT and MOAT with SKOS.

In the second case, it is possible to exploit knowledge

extraction & discovery methods like, for instance,

FFCA (see Section 3.2) that is able to conceptual-

ize a set of full-text contents and automatically clas-

sify them with respect to the extracted concepts. It

is also possible to recommend some of the connected

resources to the learner, on the basis of his/her pro-

ﬁle (Gianforme et al., 2009) in order to foster his/her

discovery process.

A sample result of this process is depicted in

Fig. 3 where the FFCA algorithm extracts the con-

cept Dublin Core from two contents, i.e., docu-

ment d and post p. The relation among the two

contents and the extracted concept is explicitly as-

serted (in the RDF/RDFS/OWL sense) by means of

the sl:isSubjectOf property. The third case fore-

sees the application of the aforementioned methods,

whose results are supervised and revised by means of

human interventions.

An important aspect of the proposed approach is

its capability to link concepts belonging to differ-

ent classiﬁcation ontologies (cross-ontologies links).

Several methods can be exploited to discover cross-

ontologies links. One of the simplest ones is to calcu-

late the document-oriented word similarity measure

(Terra and Clarke, 2003) among the labels which de-

scribe two different concepts:

rel(c

, c

) =

d f

max

. (1)

In the equation 1, d f

is the number of con-

tents annotated with both concept c

and concept c

UnlockingSerendipitousLearningbyMeansofSocialSemanticWeb

289

(coming from two different SKOS ontologies, i.e., in-

dividuals of skos:Concept). Whilst N

max

is the total

number of considered contents. An instance of the

property skos:relatedMatch is asserted among c

and c

if rel(c

, c

) is bigger than a given threshold.

3.5 Using SPARQL to Access Contents

and Concepts

The proposed approach enables high interoperabil-

ity and integration by laying upon the Semantic Web

stack. This capability allows to navigate on the on-

tology graphs by using the SPARQL language and

obtain useful results (able to concretely realize the

scenario depicted in Section 3.4) by means of simple

queries, like the following one.

. . .

SELECT ? s ? o b j

WHERE

{

? s 1 s ko s : r e l a t e d M a t c h u n i s a :DMS .

? s 1 s l : i s S u b j e c t O f ? ob j .

? s s l : i s S u b j e c t O f ? o b j

}

. . .

The previous query (it is only a sample code) al-

lows to retrieve all concepts related (in the example

we consider only the cross-ontology links) with DMS

and all contents (the result parameter is ?obj) anno-

tated with these concepts. The retrieved contents are

further analysed in order to found other relevant con-

cepts (the result parameter is ?s). This is the way to

discover that Dublin Core follows the principles of

the Semantic Web (see the scenario described in Sec-

tion 3.4).

Lastly, the inference capabilities provided by the

ontology-based reasoners (supporting RDFS, OWL

and OWL2) generate new facts (from existing ones)

which increase the knowledge base and, conse-

quently, enrich learners’ exploration experience.

3.6 Identifying Relevant Paths: A Data

Mining Approach

Suggesting hidden connections (among concepts and,

conseuently, resources) is one of the main chances

to provide surprise to learners during the exploratory

process. The method we propose is based on the

Context-Dependent Sequential Pattern Mining algo-

rithm. The idea of traditional Sequential Pattern Min-

ing (Pei et al., 2004) applied to the domain of this pa-

per is that the exploratory paths of learners, in terms

of visited concepts (on the classiﬁcation ontologies),

can be analysed in order to ﬁnd frequent sequences of

visited concepts. For instance, the algorithm can as-

sert something like this: learners who visit concepts

, c

and c

also visit concept c

. This rule can be

used to recommend c

to those learners who have vis-

ited c

, c

and c

. Moreover, the authors of (Rabatel

et al., 2013) add context awareness to the extraction

of frequent sequences. In brief, they suggest ﬁnding

frequent sequences in speciﬁc contexts. For the aim

of this paper, we can deﬁne the context as the prior

knowledge (or/and competencies) of learners. For in-

stance, it is possible to locate frequent sequences for

expert java developers, novice project managers and

so on. This approach requires the deﬁnition of rules

able to assign individual learners to one or more con-

texts. During exploration, learners could receive sug-

gestions like, for instance, resource annotated with

concept c

could be useful for you on the basis of:

i) known contextualized frequent sequences, ii) con-

texts associated to them, and iii) concepts visited in

the current session.

4 LEARNING DOMAINS AND

SCALABILITY

One of the main advantages of the Social Semantic

Web platform is that we may contextualise them for

different types of communities acting in different en-

vironments and having different sizes. This charac-

teristic is enabled by the numerous existing vocabu-

laries covering different domains. For instance, if the

community we would like to consider is composed by

the employees of a Knowledge-Intensive Organiza-

tion, we need to integrate vocabularies and ontologies

for describing projects, competencies, roles, strate-

gies, tasks, documents, training material and so on.

Otherwise, if the community we would like to con-

sider is composed by consumers related to the world

of cultural heritage, we need to consider ontologies

like CIDOC-CRM

, Geonames

, etc. Thus, if we

change the domain, Semantic Web mechanisms allow

us to change the vocabularies and ontologies but not

the approach (SKOS-based) presented in Section 3.4.

Furthermore, the Social Semantic Web platform

is highly and natively scalable. Both horizontal and

vertical scalability can be considered. More in de-

tails, horizontal scalability occurs when more than

one communities are considered. This case is il-

lustrated in Fig. 4. A community can be repre-

http://www.cidoc-crm.org

http://www.geonames.org

CSEDU2014-6thInternationalConferenceonComputerSupportedEducation

290

sented by a set of SKOS concepts schemas, a set

of persons and a set of vocabularies and ontolo-

gies modelling data (coming from software applica-

tions used in that community) which are relevant in

a speciﬁc domain. The proposed approach and, in

particular, skos:relatedMatch, skos:broadMatch,

skos:narrowMatch properties can link SKOS con-

cept schemas belonging to different communities and

enable serendipitous learning process among different

communities. Adding a new community is allowed

also at run-time.

Figure 4: Horizontal scalability.

Vertical scalability is allowed by the capability of

Semantic Web to add new information systems and,

thus, new data at run-time.

Lastly, a third type of scalability may be consid-

ered: infrastructure scalability. This means the capa-

bility to handle different and distributed storage sys-

tems for ontologies and, in general, semantic data is

guaranteed by the distributed nature of RDF.

5 EARLY EVALUATION, FUTURE

WORKS AND FINAL REMARKS

The work proposes an approach to exploit the So-

cial Semantic Web platform to support Serendipitous

Learning. The SIOC framework and its food chain

have been generalized in order to support multiple do-

mains. Two interconnected ontology types have been

introduced. The ﬁrst one is used to describe rele-

vant resources. The second one is used to classify

and connect heterogeneous resources and it is mainly

based on SKOS. Moreover, suitable methods for link

and path discovery, leveraging on classiﬁcation on-

tologies, have been proposed. The scalability of the

proposed approach has been shown. An early partial

evaluation has already been executed in the context of

the experimentation of an integrated workplace learn-

ing system (developed in the context of another R&D

Project by the same authors). More in details, we de-

ployed SMW+ (Semantic Media Wiki Plus)

to allow

users to interact with the ontology layers for a ﬁxed

time interval in order to experience and answer to a

Likert-based questionnaire. In particular, 20 workers

of CRMPA (Centre of Research in Pure and Applied

Mathematics)

, involved in four R&D Projects, were

asked to use the deployed system for two months in

order to support their project tasks. At the end of this

period, the 20 workers are asked to answer the ques-

tionnaire consisting in 12 questions. Among these

questions, the item Q10 asked workers to evaluate

their serendipitous learning experience with the de-

ployed system. In a scale from 1 (worst) to 5 (best),

the median was 4, the mode was 5, the variability was

1.2 (for range) and 3 (for iq range). All these values

refer to the item Q10. Of course, the executed exper-

imentation is largely insufﬁcient to evaluate the pro-

posed approach but it has been performed in order to

understand its potential usefulness. Thus, we have al-

ready planned to execute a complete experimentation

in a larger community of users in the context of the

SIRET Project (see Section 5), where we will set two

groups (an experimental group and a control group) of

learners (with the same objectives) and we will try to

analyse the differences in the knowledge acquired in

a group rather than in the other one. Discriminate the

knowledge acquired by means of serendipitous pro-

cesses will be one of the main challenges of the ex-

perimentation methodology. Additional experimen-

tation phases will be executed to evaluate the algo-

rithms proposed to construct and manage ontologies

and to perform link discovery and context-dependent

sequential pattern mining.

ACKNOWLEDGEMENTS

Partially supported by the Italian Ministry of Univer-

sity and Research under the PON 01 03024 SIRET,

Recruiting and Training Integrated System, project

approved by D.D. 721/Ric 2011, October 14 and

started in 2011, November 1. Special thanks to the

other partners involved in these activities: CRMPA,

Centro di Ricerca in Matematica Pura e Applicata,

Fisciano (SA), Italy and MOMA S.p.A., Baronissi

(SA), Italy.

REFERENCES

Andr

e, P., Teevan, J., Dumais, S. T., et al. (2009). Discovery

is never by chance: designing for (un) serendipity. In

http://www.smwplus.net/

http://www.crmpa.it/

UnlockingSerendipitousLearningbyMeansofSocialSemanticWeb

291

Proceedings of the seventh ACM conference on Cre-

ativity and cognition, pages 305–314. ACM.

Baker, T., Bechhofer, S., Isaac, A., Miles, A., Schreiber, G.,

and Summers, E. (2013). Key choices in the design

of simple knowledge organization system (skos). Web

Semantics: Science, Services and Agents on the World

Wide Web.

Boj

ars, U., Breslin, J. G., Finn, A., and Decker, S. (2008).

Using the semantic web for linking and reusing data

across web 2.0 communities. Web Semantics: Science,

Services and Agents on the World Wide Web, 6(1):21–

28.

Boyd, D. (2010). Streams of content, limited attention: The

ﬂow of information through social media. Educause

Review, 45(5):26–28.

Breslin, J. G., Decker, S., Harth, A., and Bojars, U. (2006).

Sioc: an approach to connect web-based communi-

ties. International Journal of Web Based Communi-

ties, 2(2):133–142.

Breslin, J. G., Passant, A., and Decker, S. (2009). Social

Semantic Web. Springer.

Brooks, C., Bateman, S., Greer, J., and McCalla, G. Lessons

learned using social and semantic web technologies

for e-learning.

Capuano, N., Dell’Angelo, L., Orciuoli, F., Miranda, S.,

and Zurolo, F. (2009). Ontology extraction from ex-

isting educational content to improve personalized e-

learning experiences. In Semantic Computing, 2009.

ICSC’09. IEEE International Conference on, pages

577–582. IEEE.

Gaeta, M., Mangione, G., Miranda, S., and Orciuoli, F.

(2013). Adaptive feedback improving learningful con-

versations at workplace. pages 175–182.

Gaeta, M., Orciuoli, F., Paolozzi, S., and Salerno, S. (2011).

Ontology extraction for knowledge reuse: The e-

learning perspective. Systems, Man and Cybernetics,

Part A: Systems and Humans, IEEE Transactions on,

41(4):798–809.

Gianforme, G., Miranda, S., Orciuoli, F., and Paolozzi, S.

(2009). From classic user modeling to scrutable user

modeling. pages 23–32.

Gritton, J. (2007). Of serendipity, free association and aim-

less browsing: do they lead to serendipitous learning.

Retrieved March, 20:2012.

Heinstr

om, J. (2007). Psychological factors behind inci-

dental information acquisition. Library & Information

Science Research, 28(4):579–594.

Ihanainen, P. and Moravec, J. (2011). Pointillist, cyclical,

and overlapping: Multidimensional facets of time in

online education. The International Review of Re-

search in Open and Distance Learning, 12(7):27–39.

Jeremic, Z., Jovanovic, J., and Gasevic, D. (2013). Personal

learning environments on the social semantic web. Se-

mantic Web, 4(1):23–51.

Jovanovic, J., Gasevic, D., and Devedzic, V. E-learning and

the social semantic web.

Kop, R. (2012). The unexpected connection: Serendipity

and human mediation in networked learning. Educa-

tional Technology & Society, 15(2):2–11.

Lu, C.-J. (2012). Accidental Discovery of Information on

the User-Deﬁned Social Web: A Mixed-Method Study.

PhD thesis, University of Pittsburgh.

Maio, C. D., Fenza, G., Loia, V., and Senatore, S. (2011).

Hierarchical web resources retrieval by exploiting

fuzzy formal concept analysis. Information Process-

ing & Management, (0).

Mangione, G. R., Orciuoli, F., Gaeta, M., and Salerno, S.

(2012). Conversation-based learning in the social se-

mantic web. Journal of e-Learning and Knowledge

Society-English Version, 8(2).

McCay-Peet, L. and Toms, E. G. (2010). The process of

serendipity in knowledge work. In Proceedings of the

third symposium on Information interaction in con-

text, pages 377–382. ACM.

Passant, A. and Laublet, P. (2008). Meaning of a tag: A col-

laborative approach to bridge the gap between tagging

and linked data. In LDOW.

Pei, J., Han, J., Mortazavi-Asl, B., Wang, J., Pinto, H.,

Chen, Q., Dayal, U., and Hsu, M.-C. (2004). Mining

sequential patterns by pattern-growth: The preﬁxspan

approach. Knowledge and Data Engineering, IEEE

Transactions on, 16(11):1424–1440.

Rabatel, J., Bringay, S., and Poncelet, P. (2013). Min-

ing sequential patterns: a context-aware approach. In

Advances in Knowledge Discovery and Management,

pages 23–41. Springer.

Scardamalia, M. and Bereiter, C. (2006). Knowledge build-

ing: Theory, pedagogy, and technology. The Cam-

bridge handbook of the learning sciences, pages 97–

115.

arez-Figueroa, M. C. (2010). NeOn Methodology for

building ontology networks: speciﬁcation, scheduling

and reuse. PhD thesis, Informatica.

Terra, E. and Clarke, C. L. (2003). Frequency estimates for

statistical word similarity measures. In Proceedings of

the 2003 Conference of the North American Chapter

of the Association for Computational Linguistics on

Human Language Technology-Volume 1, pages 165–

172. Association for Computational Linguistics.

Torniai, C., Jovanovic, J., Gasevic, D., Bateman, S., and

Hatala, M. (2008). E-learning meets the social seman-

tic web. In ICALT, pages 389–393. IEEE.

Weller, K. (2010). Knowledge representation in the social

semantic web. Walter de Gruyter.

CSEDU2014-6thInternationalConferenceonComputerSupportedEducation

292