Pattern-Based Cross Media Social NetworkAnalysis for Technology Enhanced Learning in

Europe

Ralf Klamma, Marc Spaniol, Yiwei Cao, and Matthias Jarke

Lehrstuhl Informatik V, RWTH Aachen University, Germany{klamma|spaniol|cao|jarke}@i5.informatik.rwth-aachen.de

Abstract. It is extremely challenging to get an overview of the currentstate-of-the-art in technology enhanced learning in Europe. Rapid tech-nological and pedagogical innovations, constantly changing markets, avivid number of small and medium enterprises, complex policy processes,ongoing political and societal debates on the pros and cons of technologyenhanced leaning, combined with many languages and different cultures,make it almost impossible for people to be informed. We want to in-troduce the media base and the measure tools for pattern-based crossmedia social network analysis, created by the PROLEARN network ofexcellence in professional learning. The main goal of this endeavour is thereduction of complexity for actors in digital social networks by applyingideas from social software and already successful methods for complexityreduction, such as information visualization, social network analysis andpattern languages.

1 Technology Enhanced Learning in Europe

Networking is one of the buzzwords for the information and knowledge society inthe 21st century. Our spheres of life are networked, especially with digital mediahaving impact on all forms of human action. According to the Lisbon Strategy,Technology Enhanced Learning is one of the major goals of the European Com-munity, which should also be reached by access to ICT, networks and resources.Due to the different languages in Europe, the market of the enhanced learn-ing has a larger diversity than the USA. The development levels of technologyenhanced learning vary from country to country. Above all, the United King-dom, the Netherlands and Sweden play the leading roles in enhanced learningin Europe [31]. The reasons are software industry and the high internet pene-tration rates. The United Kingdom dominates the software sector with 29.6%of the market share (EUR 43.8 billions). Statistics show that internet usage byindividuals in Sweden (82%), Denmark (76%) and Finland (70%) reached thehighest levels in Europe in the first quarter of 2004 [47].

Through this continuing process new network phenomena occurs, such asdistributed learning repositories [12], virtual learning communities and digitalnetworks, such as the European School Net (www.eun.org/portal/index.htm)and the PROLEARN Virtual Competence Center (www.prolearn-online.com),offering services for education in general or to businesses. As there are many

2 EC-TEL 2006

pedagogical innovations in technology enhanced learning and a growing num-ber of technical infrastructures, the demand for technology enhanced learningservices is growing, too. Due to the extending research on economic feasibility,commercial relevance, usability, sociability, and educational benefits, the overallsituation of the European technology enhanced learning market is getting morecomplex. Information in these networks, is far beyond the human capabilities ofanalysis and understanding, due to the sheer mass of contents, the number ofcompanies, as well as the different languages, cultures, and markets.

This paper introduces one of the endeavours of the PROLEARN Networkof Excellence in Professional Learning (www.prolearn-project.org) [51]. Inthe scope of the work package “Social Software”, we aims to create a Europeanmedia base, containing digital information about technology enhanced learningfor different target audiences, i.e. actors in digital media networks, includingscientists, business people, policy makers, and normal citizens who are interestedin ongoing discourses in technology enhanced learning. We store informationfrom mailing lists, newsletters, blogs, rss feeds, web sites, podcasts, etc. ThePROLEARN media base is far from being complete. Like any kind of socialsoftware, the media base depends on the participation of the people using thesoftware. Being started with a excellence network of hundreds of partners is goodfor having a critical mass of content. However, the barriers of participation haveto be very low [6, 45]. Therefore, the tools should be used easily and offer addedvalue services. We also aim to provide actors self-monitoring tools to help thembetter understand the impact not only of their own activities, but also of theother actors’ activities in the digital media network. By now, we do not knowdefinitely what result the idea of a European media base for technology enhancedlearning will lead to. What we know by now is that we need to reduce complexity.So the toolset PROLEARN Measure offers a series of computer-based tools foranalysis of the media in the PROLEARN media base.

Computers are deployed to facilitate our understanding of complex networkphenomena. A network is a set of autonomous entities, connected in a specificway. Natural or artificial networks share a set of entity-, sub-network or network-specific features which could be observable empirically sometimes. These featurescan be expressed by empirical measures or probabilistic models. With the com-mon mathematical basis, it is possible to exchange knowledge about networksbetween different theoretical or observation contexts. In contrast to explanations,gained in isolated scientific disciplines such as sociology, genetics or economics,this leads to a dramatic increase in the appreciation of network phenomena.

In the following we introduce three approaches to complexity reduction: in-formation visualization, (social) network analysis, and pattern languages. Allthree approaches are interwoven. The sometimes subtle interplay between visu-alization and network analysis respectively pattern presentation has to be takeninto account, too [29].

Information visualization [3] is certainly a promising way to reduce complex-ity, but we have to go beyond visualization, since the amount of information ischallenging even for advanced information visualization strategies [43]. A survey

EC-TEL 2006 3

we have conducted in 2002 shows that a particular requirement of not technicalexperienced users was of special interest in a sophisticated cross media analy-sis with hidden computational complexity. The visualization of networks withgenerally available algorithms creates (sometimes even undesirable) evidence forabstract numerical correlations. Those network representations are one more dig-ital medium in the trace of media usage. Re-contextualizing them in scientificand societal discourses facilitates knowledge creation, since societal processesand structures are mediated over digital media. The importance of digital mediafor our life is constantly growing while trust in digital media is an issue. On theone hand, digital media offer actors in networks new possibilities to act. On theother hand, these actors themselves are more often a target of undesired actionsof others. This relation of agency and patienthood is most critical in our appro-priation of information technology at large, e.g. in vocational training situationspeople are very unsure about the future. Challenges are to balance structuralanalysis and content-related analysis of media. Both content-based and struc-tural analysis can be tackled by media-extended social network analysis.

This second way to reduce information complexity is to apply social networkanalysis (SNA) [10]. Within structural traditions in sociology, SNA is applied tohandle large complex relational data sets, covering information about social re-lations. Working on this idea, we want to present cross media extensions arguingthat actors act through the media which they produce and consume. The basisfor cross media social network analysis is the Actor-Network Theory (ANT) [30].It provides an approach which does not distinguish between people and objects.For that reason, it is an appropriate method for modeling digital social networks.

The third approach to reducing complexity are pattern languages. A patternincludes the description of an observable and known behaviour related to themedia base. We later call this a disturbance. From the tradition we take ele-ments like the context where the pattern is applied, forces which come into play,the recommended solution and the rationale behind it [1]. Patterns are mostuseful if they are seen in relations with each other [42], since the human mindis always searching for regularities and similarities [41]. Patterns were alreadysuccessfully applied in other areas to reduce complexity, such as architecture [1],economics [11], and computer science [18]. A pattern language might be crucialfor forecasting the possible development of digital media. Since, in the momentthe formalization and applicabiliy of patterns are most urging for the media base,we discuss differences and similarities with other pattern language approachesin a later phase of the project.

The rest of the paper is organized as follows. In the next section we willintroduce the cross media social network analysis theory. Then, we will presentthe PROLEARN media base. On top of the media base there are a series ofexploratory and interactive tools, PROLEARN Measure. After that pattern-based methods to combine structural and content specific analysis on large datasets are presented. The paper concludes with an outlook.

4 EC-TEL 2006

2 Cross media Social Network Analysis

Cross media Social Network Analysis creates evidences about acting and beinga target of actions in media networks. These evidences are starting points forlearning processes within the networks to develop competences of acting in anetworked world or on a global stage. We are aware that all tools presentedhere can be used to observe the online behaviour of people. Therefore, it is mostimportant that the media base and the tools presented here are not in the handsof a few but are given to the network members for self-monitoring. For the sakeof clarity we introduce the theories influencing our cross media social networkanalyis theories in short.

– The structural properties of dynamic, evolving and real-world networks arepart of (social) network analysis [36].

– Media operations have been studied mainly in the humanities [21] while aunifying view on media agents and patients is possible with Actor-networktheory [30]. The structural dependencies between actors can be ex-pressed best with the i*-Framework [52].

– Learning processes based on reification of intuition, shared practice andbeliefs are studied in the context of Knowledge Management and Or-ganisational learning [38, 50].

Every network supports certain types of media. They influence how the commu-nication links between the members are created. The members as edges and thecommunication links as vertices define a graph G = (V, E). Properties whichhave their roots in the network analysis can be defined for the network and themembers. The most important network properties are the small world [49] prop-erty and the scale free network [2] property. In those networks that are bothsmall world and scale free, there exist special members, so-called hubs whichcontrol media opeations being a well-connected node. We model this within thei*-framework [52]. Apart from the network, there are three special types of ac-tors that compose a social network. The member stands for an existing personor a sub-community. The medium enables the members to do certain activities,among which the most important ones are establishing communication links andexchanging information. The artefacts are objects created by the members, us-ing some medium. The communication links among the members can be tracedusing the artefacts. The basic social unit, which forms a digital social networkis the member. Members have properties which are derived from SNA, describetheir positions, and are used to estimate the importance of the members in thenetwork modeled than in the i*-framework.

Some important types of centrality are: the Degree centrality, a propertywhich measures in the most straight forward way the member’s social capital inthe network ; the Closeness centrality CCm, taking into account the closenessof the member m to every other member [4, 10, 15]; the Betweenness centrality,estimating the possibility of a member to influence the communication of others[4, 10, 15]. Efficiency is used to detect structural holes in the neighbourhoodof a member v. Structural hole is a relationship between two non redundant

EC-TEL 2006 5

Pre-texts

TranscriptInter-medial Transcription

Understanding

and Critique

Fig. 1. Transcription creates interpretations of media in media

neighbours of a member [7]. The more structural holes exist for a member, thegreater his social capital is [19]. The efficiency is a function of the time and theenergy which a member invests into the relations with each neighbour.

Referencing ANT, we apply SNA on networks built by media, artefacts andindividuals. All media are social entities. A media-specific theory helps us un-derstand digital media support for cross media SNA. It is based on the followingmedia operations [14, 21]. New concepts are developed through making selectionsfrom an initially unstructured set of media-based ’pre-texts’ and converting themthrough a process called intra- or inter-medial transcription into a so-called tran-script, a media object describing the new concept.• It condenses and structures the set of pre-texts by designating some of them

as evidences, counter-examples, or points of critique, while de-emphasizingothers. Transcription is a media-dependent operation to make media collec-tions more readable (cf. Figure 1).

• Thus, it enables a new reading of the pre-texts where the kind of reader-ship is determined by the media, through which the transcript is preparedand communicated. Well-designed transcripts can significantly increase thecommunity of practice for a certain piece of knowledge, or may intention-ally focus on specific ’insiders’. Localization means an operation to transferglobal media into local practices.

• The term of (Re-)Addressing describes an operation that stabilizes and op-timizes the accessibility in global communication.

Thus, transcription not only changes the status of passive understanding byothers, but also enables further cooperative knowledge creation by stimulatingdebates about the selected pre-texts and about the transcription itself. More de-tails are in [22, 23]. We will now synthesize these media-specific operations withlearning processes by human actors. The result is a media centric re-formulationof the previously introduced media operations on knowledge creation and sociallearning processes adopted from Nonaka and Takeuchi [38] and Wenger [50]. Theknowledge creation theory, especially the SECI model, by Nonaka [38] has beenwidely acknowledged in management theory and practice. Also in the fields ofCSCL and professional learning, the most prominent knowledge managementtheories are those of Bereiter, Engestrom and Nonaka [39]. We are buildingour media base on the assumption that digital social networks, informal sub-networks or communities of practice want to share knowledge about business,markets, their professions etc. Knowledge sharing in this setting is primarily asocial process [9, 46, 20]. The SECI model makes a basic distinction between tacitand explicit knowledge [40, 38, 35, 26, 27]. The processes include the related me-dia specific operations: transcription, localization, and addressing. In the upper

6 EC-TEL 2006

section we focus on actions performed by actors. Starting with an individual whohas internalized some media-specific knowledge, there are two ways to communi-cate with others. On the one hand, there is an option to present this informationto others by localized actor-actor interaction, which allows the content’s social-ization within the Community of Practice and is equivalent to the development ofa shared history vice versa. On the other hand, individuals may also perform anactor transcription of their knowledge by generating new medial artefacts. Thisoperation brings us into the lower section, where media from the digital socialnetworks are processed. The externalized artefacts of a member are now furtherprocessed by the IS. This is done by a transcription of the medial artefacts.The final addressing closing the circle is the context depending presentation ofthe medial artefacts or a cross media concatenation. From then on, the processmight be repeated infinitely, oscillating between tacit and explicit knowledge onthe epistemological axis and between actors and the digital social network onthe ontological axis.

Moreover, we want to address two other issues: (a) the prerequisites for anyknowledge transformation and (b) a phase model for knowledge creation. Thisphase model of learning (sharing tacit knowledge, creating concepts, justifyingconcepts, building an archetype, cross-leveling of knowledge) makes it clear thatlearning is an action-oriented process. The aim of knowledge creation is to buildsomething (the archetype). Thus, the PROLEARN media base needs to definea mission, a vision which brings the members together. The overall goal of thePROLEARN Network of Excellence, the PROLEARN outreach instruments, thePROLEARN Academy and the PROLEARN Virtual Competence Center is es-sential for the creation of a mission. With the cross-leveling of knowledge, theprocess starts again. As important as the process model are the prerequisites oflearning, which are intention, autonomy, fluctuation, creative chaos, redundancyand requisite variety. All these ideas might amount to a nightmare for westernstyle organizations. However, Nonaka and Takeuchi are arguing that these pre-requisites are inevitable. Furthermore, the concept of ’ba’ [37] and the conceptof CoP [50] are quite similar.

3 A Technology Enhanced Learning Media Base for Europe

The first question addressed is why to build up a media base. Briefly, this enablesus to perform a model-based analysis of different kinds of processes in existingdigital social networks to describe and relate concepts of communication, coordi-nation, and knowledge organization, compare and extract use of these conceptsin different contexts, identify patterns of communication and knowledge orga-nization, and study common and distinct features of digital media usage. Animportant feature is to offer members a functionality to collaboratively admin-istrate and retrieve medial artefacts. Of course, there are many projects dealingwith thematic classification of artefacts for a latter retrieval. For instance, thereare several social software tools indexing and classifying bookmarks, references,images, videos in general. We analyze the correlation between digital social net-works and the media they use. For that purpose, we support the members by

EC-TEL 2006 7

COMMUNITY USER

isA

belong

rank

belong

belong

create

belong

address

consist

SUB_LC

LEARNINGCATEGORY

consist

1

10

CATEG

isA

MEDIUM

consist

has

M_TYPE

isA

p

PROCESS

consume

has

affectedby

perform

n

P_TYPE

isAacquisition

searchtranscription

retrieval

monitoringp

ACTOR

has

consist

A_TYPE

isAconsumer producerp

p

n 1

1

n

1 n

1

n

n m

n m

1 n

1

n

m n

m n

1

n

mn

mn

p

n

1

1

n

admin viewerdigital non-digital

Fig. 2. ER Diagram of the Community centered PROLEARN Media Base

tools to classify, comment and review the projects they monitor. In this aspect,projects in the cultural sciences cover discussion groups, newsletters, web-sites,feeds, and blogs.

The second question addressed is how to build up such a media base. Wedecided to build the core of the media base on top of an operational commu-nity memory system, called GRAECULUS [27]. The strategy was preferred,because the system already interrelates patterns of knowledge organization andpatterns of communication and coordination. As cited in [6], we are membersof hundreds of different overlapping communities all the time. To interrelate theknowledge organisation processes with the community processes, we constructan Entity-Relationship model that emphasizes the coherence between the com-munity process models and the knowledge organization processes. The ER modeldepicted in Figure 2 is based on an extended entity-relationship approach. Eachbox represents an entity of the model, with the attributes omitted here. Relation-ships are linked by diamonds with links to the entities, while isA-relationshipsare presented as triangles. The most important concepts of the ER model aresummarized in the next paragraph.

A Community is the central concept of the PROLEARN media base, asit is in the middle of a whole network and the individual members. In everykind of social software, there are means to facilitate such sub structures. With-out them, social software would not be usable any more because of the highcommunication complexity and trust issues. Community activities contain im-portant knowledge, which is stored on media belonging to the community. Ac-tors also belong to Communities and perform processes (modeled by processmodels expressed in a language chosen by the actors). The Process conceptdescribes community and knowledge processes of actors. All Processes can berefined by consist-relationships. Each process belongs to a process type taxon-omy (P TYPE ) which is defined collaboratively by each community. We define astandard taxonomy [42, 48] with five different process types: a) acquisition pro-

8 EC-TEL 2006

cesses which are used to acquire new media, b) search processes for media, c)transcription processes for media, d) retrieval processes for media, and e) moni-toring processes for media. An Actor characterizes humans or groups of humans(consist-relationship) performing a process. Each actor belongs to an actor typetaxonomy which is defined collaboratively by each community. From past ex-periences with knowledge management projects [34], we have already defined astandard taxonomy consisting of four different actor categories: a) Actors whoproduce media, b) Actors who consume media, c) Actors who view media, andd) Actors who administrate or manage media within a community. The Mediumconcept describes artefacts created or consumed by the processes, according tothe transcription theory. For later use in cross media social network analysisand in the pattern language, we can differentiate between media and artefacts.But this is not shown here. Each Medium belongs to a media type (M TYPE ),describing a taxonomy of media types which is also defined collaboratively byeach community. ANT states that actors not only perform processes, but alsoare affected by processes (patienthood). A User is a special actor that uses thePROLEARN media base. Therefore, the User has a certain rank within the com-munity. Each Community has a classification. The classification can be done by’tagging’ of the community members or can be defined in other categorizationhierarchies. Here, we put a learning classification developed by the PROLEARNnetwork of excellence to characterize their work in work packages. Other classi-fications which can also be used in parallel are geographical distributions of themedia and language distribution etc.

Retrieval and Exploration ToolsRetrieval of data is realized on top of the content management system Plone,which builds on the application server ZOPE. For each medium in the mediabase, we have created web-based access structures, e.g. for reading mails, newslet-ters, blog entries, rss feeds and so on. All web sites in the media base are mirroredcompletely on local file systems, so we can access them even if the original website is not available temporarily or gone forever. For accessing the huge amountof information, we have predefined - apart from a simple free text search function- several queries that enable even a not experienced user to discover dependen-cies among various media offers in order to explore its content by the structuredefined in the ER diagram. First, users might pass through the media structureand its components. Here, various characteristics of the projects are displayed,which have been automatically extracted upon its creation and can be modifiedmanually later on. Another way is to deploy information visualization strategies.

Figure 3 shows a treemap visualization of the PROLEARN Measure mailinglists. The mailing lists are categorized according to their languages and accordingto some categorization schemes on learning. The latter is derived by adoptingsome of the PROLEARN work packages as the categories. The treemap is com-puted by using the squarified treemap algorithm [5]. The size of the rectangleson the lowest level reflects the number of mails that have been sent within thecorresponding mailing lists. For each mailing list, some statistics, such as theoverall number of mails, can be displayed. In order to provide a close look at the

EC-TEL 2006 9

Fig. 3. Treemap Visualization of the PROLEARN mailing lists

mailing lists, it is possible to zoom in the treemap. Further development aimsat visualizing the mails themselves as well, i.e. going to one level deeper.

In Figure 4 a simple tool for the cross media social network analysis of thePROLEARN Measure mailing lists is shown. The user can select different net-works, explore them and perform an interactive cluster analysis. In the following,this process is depicted in a more detailed way: at first, the user has to selectthe mailing list he wants to explore and a time frame for this exploration (seeform “Data” in Figure 4). Then the user has to decide on the nature of the datathat he wants to visualize as a network.

In this context, four different kinds of information (and arbitrary combina-tions of them) can be displayed: 1) If the option “Mails Member ->Member” ischosen, this results in a network reflecting the communication structure betweenthe members of the selected mailing list. More precisely, if the member A hasreplied to a mail by the member B, the network contains an edge from A to B.2) With the option “Mails Member ->Mailing List” turned on, each member towhom A has written a mail in the selected mailing list, results in an edge fromA to a node representing this list. The last two options concern the referencestructure of the mailing list: 3) If some web site W has been referenced in one ofthe mails by the member A, the resulting network contains an edge from A to W(third option). 4) Basically the same as 3), only that all mails of the mailing list(also those that have been sent by the mailing list itself) are considered. In thiscase, for each mail in which the web site W is referenced, the network containsan edge from the node representing the mailing list to W. If the user selectsseveral options at the same time, the corresponding networks are merged.

After having submitted the “Data” input, the final network is displayed.The selected mailing list is represented by a mail icon, an individual by an ac-tor icon and a website by a globe icon. This network can be further exploredthrough the functionalities of zoom-in and zoom-out, moving it within the net-work window, moving single nodes or groups of nodes and choosing different lay-

10 EC-TEL 2006

Fig. 4. Cross media Social Network Analysis for PROLEARN Measure

out algorithms. Four different layouts are available: self-organizing map by Meyer[33], Fruchterman-Reingold [17], Circle, and Kamada-Kawai [25]. As mentioned,apart from simply exploring the network, it is possible to cluster it interactively.The clustering is based on the Edge Betweenness Centrality (EBC) [16]. With aslider, the user can choose the number of edges N he wants to be removed fromthe network. This results in the removal of the N edges with the smallest EBC.The different colours of the nodes then indicate the latter’s cluster membership.When clicking the button “Group Clusters”, all nodes that belong to the samecluster are grouped together.

4 Pattern-based repositories for technology enhancedlearning media

Our PALADIN (PAttern LAnguage for DIsturbances in digital social Networks)approach is an application, developed to research into a set of digital social net-works with patterns on the basis of the ANT model and a pattern language.We call behaviour observed in digital social networks behaviours because it isthe start of a learning process triggered by the difference between the actualbehaviour and the desired behaviour. PALADIN offers a web interface to de-fine, browse and store patterns, actors and actor properties. The repository isan XML database management system. PALADIN includes a general networkimplementation that focuses on the structural properties of the members. InPALADIN a digital social network is a set of actors along with their relations.It has properties which can be used in the definition of the patterns. A mediumor a set of media is the basis of the digital social networks. It influences the net-works, their structures and the way to communicate [44]. Dependencies betweenactors are modeled with the i*-framework [52]. Every type of medium defines

EC-TEL 2006 11

how the communication links between the members are built. The media areemails, mailing lists, blogs, transaction-based web sites, wikis, URL, and chatrooms.

The communication between members is realized through the exchange ofartefacts. An artefact is created by a member in a certain medium. The artefactsrepresent the information circulating in the digital social network. The types ofartefacts, analyzed in PALADIN, are messages, threads, bursts, conversations,blog entries, comments, web pages, transactions, and feedbacks. The propertieswhich are mostly used for the pattern definition are the author and the creationdate. Most of them are defined intuitively. An exception is the burst which is aconceptual artefact, based on the activity in the network for a given time period.It can be used to detect topics which appear, gain popularity, and then fade.

According to the activity of the members and their patterns of communica-tion, we defined several patterns based on the analyis of literature [8, 13, 32, 48]like questioners, answering persons, trolls, conversationalists, and spammers. Aquestioner is a member, who seeks help and information. A answering personprovides answers to the raised questions without going into discussions. A con-versationalist is a member who actively participates in discussions. A troll aimsat drawing attention and starting useless discussions. A spammer sends messagesirrelevant for the community, which are mainly advertisements or binary files.

The pattern language includes a pattern structure definition, the Formal Ex-pression Language for Patterns (FELP), and algorithms for the application ofthe patterns in digital social networks. A pattern has a name, a disturbance, adescription, forces, force relations, a solution, a rationale and pattern relations.The name of the pattern is short but descriptive. The disturbance is a condition,which indicates the existence of the pattern in a social network. It is defined bya FELP expression. It consists of variables and rules for constructing formalexpressions. When a pattern is defined, the variables of an expression are boundto variables from the network model. The description explains the problem towhich the pattern provides a solution and the settings when the pattern occurs.The force is the basic component of the pattern. It represents an actor relevantto the disturbance. The forces in a pattern are automatically extracted at themoment when the user binds the variable from the disturbance expression. Theforce relation corresponds to a relation between actors, included in the patternas forces. The solution contains the actions, which are proposed to be carried outin the situation, to which the pattern refers. The rationale is used for reasoningabout the forces and the disturbance. It may include examples, past success sto-ries or failures. The pattern relations show whether the pattern, being currentlydefined, has anything in common with other patterns. The pattern relationsare important for the structure of the pattern language. An example may illus-trate our approach. A troll is a person which only answer in threads started byhimself. This behaviour may be recognized as disturbance by other actor in anetwork. It depends on the existence of thread artefacts in the medium. In orderto identify the trolls of a given medium we have to create a pattern expressionwhich is then an instance of an existing pattern troll. The expression looks like

12 EC-TEL 2006

(∃[troll|(∃[thread|(thread.author = troll)∧(count[message|(message.author = troll)∧

(message.posted = thread)])> minPosts])∧(¬∃[thread1,message1|(thread1.author1 6= troll)∧

(message1.author = troll ∧ message1.posted = thread1)])])

A sample set of patterns has been defined and tested on the social networksbuilt upon the available database sources. The patterns reflect the existence oftrolls, spammers, conversationalists, questioners, answering persons, bursts andstructural holes in the digital social networks. The storage of the patterns asXML documents enables their automatic application. More details can be foundin [28].

5 Conclusions and OutlookIn this paper we have presented first realization ideas for the ongoing work onthe PROLEARN media base and the PROLEARN Measure tools for technol-ogy enhanced learning in Europe. The main goal of this endeavour carried outin the PROLEARN Network of Excellence in Professional Learning and Train-ing is to give actors an overview of current discourse of technology enhancedlearning relevant to Europe. The main idea is to reduce complexity in orderto give the control of the actions back to actors in digital social networks. Notonly the impact of their own actions (media agents), but also that of the otheractors’ actions of (media patients), should be made visible by three strategiesto reduce complexity. The first strategy is information visualization with simpleprinciples, such as ’Overview first, details later’. We have presented retrieval andexploration tools which can be easily applied within the context of the PRO-LEARN media base. A second strategy is the combination of existing socialnetwork analysis tools with cross media and knowledge management theoriesto produce new cross media analysis tools dealing with both the content andthe structure of information. A third strategy is the development of a patternlanguage. With a predefined pattern, describing problems and solutions withincomplexity reduction on a natural language level which can be understood bynon technical actors easily. We can give monitoring tools to a broad target au-dience to make possibly new interactions among them. Our hope is that newlearning processes in communities, supported by the PROLEARN media base,take place. Furthermore, this will be a never-ending circle of knowledge creation.

These are the first experiments we undertake with social software in thecontext of professional learning. The PROLEARN media base and the PRO-LEARN Measure tools are integrated in the PROLEARN Academy portal (www.prolearn-academy.org) to reach a broad audience inside and outside the net-work. We have ongoing discussions among the network members to add media tothe media base and to join forces with fellow projects, such as the EU STREPSICamp (www.icamp-project.org) [24] and the other projects from the Profes-sional Learning Cluster (www.professional-learning-cluster.org). We donot know how the media base will develop in future. As stated in [6], some ofthe best communities are not built - they emerged.

EC-TEL 2006 13

AcknowledgementsThis work was supported by German National Science Foundation (DFG) within thecollaborative research centers SFB/FK 427 “Media and Cultural Communication”, theSFB 476 “IMPROVE”, and by the 6th Framework IST programme of the EC throughthe Network of Excellence in Professional Learning (PROLEARN) IST-2003-507310.We’d like to thank the reviewers for the comments and our colleagues for the support.In particular, we thank Dimitar Denev and our student workers Heike Haegert for theimplementation.

References1. C. Alexander. A Pattern Language: Towns, Buildings, Construction (Center for

Environmental Structure Series). Oxford University Press Inc, USA, 1978.2. A.-L. Barabasi. The physics of the web. Physics World, 14:33–38, 7 2001.3. B. B. Bederson and B. Schneider, editors. The craft of information visualization.

Morgan Kaufmann, 2003.4. U. Brandes and T. Erlebach, editors. Network Analysis - Methodological Founda-

tions, volume 3418 of LNCS. Springer, Berlin Heidelberg New York, 2005.5. D. Bruls, C. Huizing, and J. van Wijk. Squarified treemaps. In W. de Leeuw

and R. van Liere, editors, Data Visualization 2000, Proceedings of the joint Eu-rographics and IEEE TCVG Symposium on Visualization, pages 33–42, Vienna,2000. Springer.

6. L. Bryant. Smarter, simpler, social. Technical report, Headshift.com, 2003.7. R. S. Burt. Structural Holes: The Social Structure of Competition. Havard Business

Press, Cambridge, MA, 1992. CSCL, SC, VC.8. K. S. Cheung, F. S. Lee, R. K. Ip, and C. Wagner. The development of successful

on-line communities. International Journal of the Computer, the Internet andManagement, 13(1):77–89, 4 2005.

9. R. Daft and R. Lengel. Organizational informations requirements, media richnessand structural design. Management Science, 32(5):554 – 571, 1986.

10. A. Degenne and M. Forse. Introducing Social Networks. Sage Publications, 2003.11. V. Dhar. Data mining in finance: Using counterfactuals to generate knowledge

form organizational information systems. Inf. Systems, 23(7):423 – 437, 1998.12. P. Dolog, N. Henze, W. Nejdl, and M. Sintek. Personalization in distributed e-

learning environments. In WWW Alt. ’04: Proceedings of the 13th internationalWorld Wide Web conference on Alternate track papers & posters, pages 170–179,New York, NY, USA, 2004. ACM Press.

13. D. Fisher, M. Smith, and H. T. Welser. You are who you talk to: Detecting rolesin usenet newsgroups. In Proceedings of the 39th Hawaii International Conferenceon System Sciences, 2006.

14. J. Fohrmann and E. Schuttpelz. Die Kommunikation der Medien. Niemeyer,Tubingen, 2004 (in German).

15. L. Freeman. Centrality in social networks. conceptual clarification. Social Net-works, 1:215–239, 1979.

16. L. C. Freeman, S. P. Borgatti, and D. R. White. Centrality in valued graphs: Ameasure of betweenness based on network flow. Social Networks, 13:141–154, 1991.

17. T. M. J. Fruchterman and E. M. Reingold. Graph drawing by force-directed place-ment. Software - Practice and Experience, 21(11):1129–1164, 1991.

18. E. Gamma, R. Helm, R. Johnson, and V. John. Design Patterns: Elementsof Reusable Object-Oriented Software. Professional Computing. Addison-Wesley,Reading, 1994.

14 EC-TEL 2006

19. M. S. Granovetter. The strength of weak ties. American Journal of Sociology,78:1360 – 1380, 1973.

20. E. Hollender, R. Klamma, D. Borner-Klein, and M. Jarke. A comprehensive studyenvironment for a talmudic tractate. DS-NELL, 2001.

21. L. Jager and G. Stanitzek, editors. Transkribieren - Medien/Lekture. WilhelmFink Verlag, Munich, 2002 (in German).

22. M. Jarke, X. T. Bui, and J. M. Carroll. Scenario management: An interdisciplinaryapproach. Requirements Engineering, (3):155 – 173, 1998.

23. M. Jarke and R. Klamma. Metadata and Cooperative Knowledge Management. InA. Banks Pidduck et al., Editors: CAISE 2002, LNCS 2348, pages 4-20, Springer,Berlin, 2002

24. B. Jerman-Blazic, T. Klobucar, T. Arh, B. Kieslinger, F. Wild, and E. L.-C. Law.An approach in provision of interoperability of elearning systems in enlarged EU -the case of ICamp project. aict-iciw, 0:6, 2006.

25. T. Kamada and S. Kawai. An algorithm for drawing general undirected graphs.Information Processing Letters, 31:7 – 15, 1989.

26. R. Klamma and M. Jarke. Driving the organizational learning cycle: The case ofcomputer-aided failure management. In W. R. J. Baets, editor, Proceedings of the6th European Conference on Information Systems (ECIS’98), Aix-En-Provence,France, vol. 1, pp. 378 – 392, Euro-Arab Management School, Granada, 6 1998.

27. R. Klamma and S. Schlaphof. Rapid knowledge deployment in an organizational-memory-based workflow environment. In H. R. Hansen, M. Bichler, and H. Mahrer,editors, Proceedings of the 8th European Conference on Information Systems (ECIS2000), Vienna, Austria, pages 364 – 371, 2000.

28. R. Klamma, M. Spaniol, and D. Denev. PALADIN: A Pattern Based Approach toKnowledge Discovery in Digital Social Networks. accepted, I-Know 2006, SpecialTrack on knowledge visualization and knowledge discovery, 2006.

29. L. Krempel. Visualisierung komplexer Strukturen: Grundlagen der Darstellungmultidimensionaler Netzwerke Campus Verlag, New York Frankfurt, 2005 (in Ger-man).

30. B. Latour. On recalling ant. In J. Law and J. Hassard, editors, Actor-NetworkTheory and After, pages 15–25. Oxford, 1999.

31. elearning - Kapital fur die Zukunft. http://www.ecin.de/state-of-the-art/elearning/, 2001 (in German).

32. T. Madanmohan and S. Navelkar. Roles and knowledge management in onlinetechnology communities: an ethnography study. International Journal of WebBased Communities, 1:71–89, 2004.

33. B. Meyer. Self-organizing graphs- a neural network perspective of graph layout.LNCS 1547, pages 246–262. Springer Verlag, 1998.

34. P. Mika. Flink: Semantic web technology for the extraction and analysis of socialnetworks. Journal of Web Semantics, 2005.

35. J. Mylopoulos, A. Borgida, M. Jarke, and M. Koubarakis. Telos - a languagefor representing knowledge about information systems. ACM Transactions onInformation Systems, 8(4):352 – 362, 1990.

36. M. Newmann, A.-L. Barabasi and D. Watts, editors. The structure and dynamicsof networks. Princeton University Press, Princeton and Oxford, 2006.

37. I. Nonaka and N. Konno. The concept of ’ba’: Building foundation for knowledgecreation. California Management Review, (40):3, 1998.

38. I. Nonaka and H. Takeuchi. The Knowledge-Creating Company. Oxford UniversityPress, Oxford, 1995.

EC-TEL 2006 15

39. S. Paavola, L. Lipponen, and K. Hakkarainen. Epistemological foundations for cscl:A comparison of three models of innovative knowledge communities. In G. Stahl,editor, Computer Support for Collaborative Learning: Foundations for a CSCLcommunity. Proceedings of the Computer-supported Collaborative Learning 2002Conference, pages 24–32. Erlbaum, 2002.

40. M. Polanyi. The tacit dimension. Anchor Books, New York, 1966.41. E. Rosch. Principles of categorization. In Cognition and Categorization, pages 312

– 322. Lawrence-Erlbaum Ass., 1978.42. N. A. Salingaros. The structure of pattern languages. Architectural Research

Quarterly, 4:149–161, 2000.43. J. Scott. Social Network Analysis - A Handbook. Sage Publications, London et al.,

2nd edition edition, 2000.44. C. Seeling, Spaniol, M., B. A., and K. R. Discourse visualization strategies for a

comprehensive medial analysis of cultural science communities. In K. Tochtermannand H. Maurer, editors, Proceedings of I-KNOW ’04, Graz, Austria, J.UCS, pages337 – 344. Springer, 2004.

45. C. Shirky. Social software: A new generation of tools. Esther Dyson’s MonthlyReport, 10, 2003.

46. E. W. Stein and V. Zwass. Actualizing organizational memory with informationtechnology. Information Systems Research, 6(2):85 – 117, 1995.

47. Compendium of projects until 2003: A world of learning at your fingertips — pilotprojects under the elearning initiative. http://europa.eu.int/comm/education/

programmes/elearning/projects_en.html, 4 2004.48. T. C. Turner, M. A. Smith, D. Fisher, and H. T. Welser. Picturing usenet: Mapping

computer-mediated collective action. Journal of Computer-Mediated Communica-tion, 10:7, 2005.

49. D. J. Watts and S. H. Strogatz. Collective dynamics of small-world networks.Nature, 393:440–442, 1998.

50. E. Wenger. Communities of Practice: Learning, Meaning, and Identity. CambridgeUniversity Press, Cambridge, UK, 1998.

51. M. Wolpers and G. Grohmann. PROLEARN: technology-enhanced learning andknowledge distribution for the corporate world. Int. J. Knowledge and Learning,1(1/2):44–61, 2005.

52. E. S. K. *Yu and J. Mylopoulos Towards Modelling Strategic Actor Relationshipsfor Information Systems Development – With Examples from Business ProcessReengineering Proc. 4th WITS, Vancouver, Canada, pages 21-28, 1994