No Guru, No Method, No Teacher: Self-Classification and Self-Modelling of E-Learning Communities Zinayida Petrushyna and Ralf Klamma Information Systems RWTH Aachen University, Ahornstr. 55, D-52056 Aachen, Germany {petrushyna|klamma}@dbis.rwth-aachen.de Abstract. Learning processes are an infinite chain of knowledge transformation initiated by human collaboration. Our intention is to analyze E-Learning communities. The current drawback of communities is a lack of common vocabularies that can be used for an E-Learning community description, design, evolution, and comparison. We examine structural and semantic parameters of E-Learning communities gathered in MediaBase of the PROLEARN Network of Excellence for professional learning. Using the parameters and the community-of-practice theory we define more standard description for a particular community or a set of communities and to identify factors that are essential for identifying overlappings between communities. 1 Introduction While the institutional context for formal learning is still dominant and supported by the European Community, e.g. in the Bologna process, the importance of self-regulated and life-long learning is becoming more and more important for a growing number of knowledge workers and people with continuing training needs. Different educational roadmaps like the PROLEARN Roadmap for Technology Enhanced Professional Learning [22] or the Open Educational Practices and Resources (OLCOS 2012) Roadmap [17]. The latter is supported by national programs like the German ”New Media in continuing education” [16] and ”Web 2.0 in continuing education” [1]. The roadmaps identify the needs of those working in demanding professional businesses where the half-life of knowledge is very short. In the future we have to learn during work time, we have to do it on-line, we have to choose our learning materials and our learning partners by ourselves and we have to re-mix them, thus becoming not only learners but also learning content authors and even teachers. We will need new competences and meta-competences for life-long learning and we will acquire it under constant external market pressures and the forces of globalization. Does not sound like a happy future for learners! On the contrary, today, we have the unique opportunity to choose learning materials from an almost unrestricted pool of resources and to choose learning partners from the many people available on the net. Learning visionaries like John Seely Brown and Richard P. Adler [6] have already transferred the concept of the long tail [2] to the realm of technology enhanced learning to make this point. We are not bound anymore to formal learning contexts like schools 2 and universities but can expand our human skills in any direction at any age. While both views, the life-long learning pressure and the bright future of unlimited learning have their own truth, we think the right way is somewhere in the middle. Even in the long tail with its millions of learning opportunities the success of learning will not only depend on the intellectual capital of learners but also on their social capital [5,8]. Social capital can be defined as the "stock of active connections among people: the trust, mutual understanding, and shared values and behavior that bind the members of human networks and communities and make cooperative action possible. (...) Its characteristic elements and indicators include high levels of trust, robust personal networks and vibrant communities, shared understandings, and a sense of equitable participation in a joint enterprise - all things that draw individuals together into a group" [8, 4]. In this paper we will undertake a journey into the long tail of learning, investigate how learning communities are structured and what they are trying to learn. Our approach is analytical, as we study data sets collected from different sources in the PROLEARN Academy MediaBase (www.prolearn-academy.org). We combine different analysis methods in a new structural-semantic way. One fine day, we want those methods in the hands of the learning communities. Our approach is constructive. We create tools for learning communities. Our approach is reflective. We want the learning communities to make their needs, their mutual dependencies, their cooperative action more explicit by modelling them on a strategic level. In the end, there are millions of learning communities out there in the long tail, every community is unique in all the components described. But our research goal is not only to allow communities a better understanding what they are actual doing in their learning niches but also to enable them to find similar communities to exchange ideas, methods and knowledge. Imagine two up to now disconnected communities about poetry from England in the Middle Ages, one in Scandinavia one in Latin America. Can they find each other? And was does it mean for them? For professional communities there will be still the need to match learning goals and acquired competences with profiles of their current or future employees. A lengthy discussion in the media about unwanted traces left in social networking sites affecting future employability of people can be turned into the systematic care of one’s portfolio for better employability. The rest of the paper is organized as follows. In the next section we introduce the basic concepts of our approach. In Sections 3 and 4 we demonstrate the technical means for self-observing and self-modelling E-Learning communities. In Section 5 we give first results from the application of our approach while a discussion and an outlook is concluding our paper. 2 Modelling E-Learning Communities as Communities of Practice E-Learning communities can be examined as a Community of Practice (CoP) [37], a community of interest (CoI) or both. The most significant difference of a community representation is that for a CoP learning is proceeding from one domain instead of a CoI that defines learning as a process within different domains. Empirical experiences from [13] show how risky it is to study and to manage learning communities modeled by a simple CoI. 3 Fig. 1. The lifelong loop ”Communities of practice are groups of people who share a concern or a passion for something they do and learn how to do it better as they interact regularly” [37].The understanding of communities from a dynamic perspective is essential, because the perspective facilitates reasoning about real objects such as complex dynamic systems that evolve over time. We consider the process of communities evolvement in a lifelong loop that includes three components: self-modelling, self-classification and disturbances as presented in Figure 1. Communities are influenced by various external factors called disturbances. The self-classification of the community is divided into two parts: the measurement and the analysis phase. The first defines a set of properties that characterizes community structures and semantics and that includes cross-media analysis. The output of the measurement phase is used in the second, the analysis phase, to build hierarchies or clusters of communities and to identify patterns. The patterns are generally repeatable solutions to commonly-occurring problems, i.e. disturbances [24]. In the self-modelling component of the loop the previously described calculations and conclusions are used so that modells can be created. As a matter of course the disturbances coming from outside change the community inside. That is the reason to re-classify and re-model communities in the lifelong loop in order to get the correct actual information about the communities. A CoP is characterized by three dimensions introduced by Wenger [37]: – Mutual engagement Community members are engaged in interaction within a community. Nevertheless, a membership in a community is not just a belonging to one organization; a CoP is not only a set of members having personal contacts with other members. – Joint enterprises Members of a CoP should negotiate communally and the CoP mediate those negotiations. These communally negotiations result in the intention that the members have. The intention is supported by different technologies, policies and rules explored by the CoP. – Shared repertoire The repertoire consists of words, tools, ways of doing things, actions and concepts that a CoP has produced. 4 For applying CoP concepts to E-Learning communities we need a theory that tries to explain social order through relations between human agents, technologies and objects. The applicable theory should model the co-working of different dimensions of a CoP, i.e. mutual engagement(ME), joint enterprises(JE) and shared repertoire(SR) that are presented not only by humans but also by non-humans actors. The actor-network-theory (ANT) [28] doesn’t distinguish between human and non-human actors and examines the networks formed by humans in collaboration with media [9]. Such an non-differenciation between people and technologies intertwines actions, influences, or results of actions. Any social action performed using technologies is influenced by these. The type of media we are using for any task affects our behavior and our position in a society. We choosed the i* framework for notating E-Learning modells. The framework stands out for opportunities to describe relations between actors in frames of a particular social system in a clear way. An i* model focuses on motives, interests, and options of an actor that plays a role in the examined system. Such modelling is appropriate for a community representation because it reflects social characteristics of system construction [38]. The framework gives an opportunity to visualize dependencies (described as ME for a CoP) and goals (JE in the CoP). Further with the help of the model we are going to follow the intentions within the communities in order to identify the users with particular behaviors, i.e. experts, novices, etc. Our view of an E-learning community system is a system with three CoP dimensions in Figure 2: Fig. 2. i* E-learning community model – Everything about members and their interactions is mutual engagement (ME) dimension. The ME is presented by an agent called ”Membership”. The meaning of an agent is described by agent’s dependencies and responsibilities available in i* Modelling. • Dependee is an agent that influences the performance of other agents. The ”Members” agent is dependent on the ”Membership” agent. The dependence between agents is presented as the task called ”Interact”. If some members of a 5 community are not interacting, their dependency between the ”Membership” and the ”Members” agent is not working thus they have no impact in the community. • The ”Learning process” agent is an unavoidable part of a learning community. A change that happens in the ”Membership” agent (ME changes) affects the ”Learning process” agent. The dependency between the ”Learning process” and the ”Membership” is presented by dependum called softgoal. Generally, the softgoal is used when the goal is unable to be described clearly, though a dependency exists and has an influence. – Joint enterprises (JE) reverberate in communities as its technologies and enterprises. Especially, they are important for E-Learning communities because the communities are strongly based on technologies, i.e. Web 2.0. Surely, the agents create a set of policies, rules or regulations within a particular community. Only members which know the community regulations are inside the community. Otherwise they are outside. Resource ”Policies & rules” is the dependum between ”Technology & Enterprise” and ”Members” agents. The JE agent, i.e ”Technology & Enterprise”, has its impact on the ”Learning Process” agent. – Shared repertoire (SR) is the knowledge of a community. The ”Knowledge” agent is the context identity of the community. The ”Members” agent should possess the context identity that correlates with the community identity. The dependum between the ”Members” and the ”Knowledge” agents called the ”Identity” resource dependum. The influence of the ”Knowledge” agent on the ”Learning process” is identified as the softgoal. With the help of the i* Framework, COP concept and the analysis we introduce different types of communities within the long tail of E-learning: 1) a question-answer community; 2) an innovation community; 3) a disputative community. The details about those models are presented in the self-classification part of the paper, i.e. Section 5. In the following section it is explained how the data was processed and metadata extracted. 3 The self-monitoring of the repository of E-Learning resources The data set we used in our experiment was created in the scope of the PROLEARN Network of Excellence supported by the IST (Information Society Technology) Program of the European Union. A part of PROLEARN is the PROLEARN Academy with its MediaBase. Its data set collects different media in professional learning, e.g. mailing lists, newsletters, feeds, websites. The data collection was organized according to the architecture of the MediaWatchers in [33] for communities in the cultural sciences and the BlogWatcher in [23]. Here we focus on mailing lists and the MailinglistWatcher. First of all a mailing list is a CoP as it possesses three necessary characteristics, i.e. interactions between members, same number of rules and regulations and the same lexicon. Moreover the mailing list is a medium with a number of threads. A thread has one or more mails (posts). Threads are identified through IDs of mails that start new discussions. The mails are called root mails. If a mail has a reference to a root mail - it is automatically in a thread of the root mail, instead if a mail has no reference, it is a root mail of a new thread. The presented approach of identifying threads based on a mail header reply_to field is not always 6 successful. The header can include none reply_to value, a reply_to value that points not to the root mail, but to the general mail of the mailing list or to the other author in the thread, etc. In order to refine thread structure we examined a subject field as well. Before analyzing the content of mailing list communities we tested its consistency. The thread content should include the text that senders wrote and posted for the others to read. It is always the case that a mail can include technical stuff (e.g. HTML, CGI, etc.) that disturbs mail content analysis. Moreover when a sender replies, her reply may include the mail to which she replies consequently the mail body. As there is no applicable standard, it is possible that a mail appears before a reply mail or after the reply mail. A thread structure is complex: a root mail can have several answers within a thread; the answers may have replies, the replies may have replies, etc. The thread is not structurally consistent while it is full of technical stuff, duplicates, a symbol or a set of symbols that has no functions for thread semantics. We avoided HTML tags and leave only text data. Other sequences of repeated characters that have no sense were successfully deleted by using regular expressions. Finally, the data refinement script uses complex algorithm that utilize the Levenshtein distance [29] and results in the complete structure of mailing lists threads. We analyzed E-Learning communities as social networks [7], that give an opportunity to focus rather on relationships between community members than on members attributes [30]. We consider a member of a community as a node of a graph G and a relation of the member to the other one as an edge of the graph. The important concepts of social networks are as follows [36]: connections between members are channels for transfer or flow of either material or nonmaterial resources. We performed semantic analysis of E-Learning communities as well. It is based on text examination. Linguistics is able to analyse and characterize the gist of text items. We used the currently prevailing approach in computer linguistics, i.e. statistical natural language processing (NLP). The metadata got after applying described analysis is differentiated according to the CoP dimensions. Mutual engagement includes the measures that attend the interactions of members. Connectiveness, biconnectiveness . A pair of graph nodes a and b is strongly connected if it is a directed path from a to b. The connectiveness property counts the number of closely connected structures of the graph G. Biconnected components are 2 nodes in the graph G where each of the node has connection the other[11]. The connectiveness properties indicate how dense the connections are within a community graph, i.e. how diverse a community is and what are the relations within items of the community. Hubs, authorities and scale free network . A scale free network possesses a lot of nodes with a minor number of edges and several nodes with a large amount of edges. The latter type of the node is called a "hub" in the network [4]. The distribution function of scale free networks obeys the power law [3]. If a network is not scale-free, all nodes are more or less the same at all scales and it is complicated to differentiate between the nodes [34]. The ”authorities” are the nodes that are pointed from many others, i.e. hubs. The nodes in a network possess hubness or authoritativeness properties [26]. Degree centrality, closeness centrality, and betweenness centrality. The study of inand out- edges of a node defines centrality properties. The higher the degree centrality of a node is, the more visible the node is in a network. Closeness centrality shows how 7 close a node is to the others [10,15,19]. Betweenness centrality identifies how many paths are going through members, how many times they are bridges of information. Emotions. The other important aspect of ME is an emotion as it is an unavoidable part of negotiations between members of communities. It reflects the full complexity of doing things together [37]. Information about moods can be classified as aggressive, supportive, with an interest, etc. [20,31]. Emotion frequency counts how many phrases from the manually constructed list of emotional phrases are present in a mail content. The list from LIWC project (http://www.liwc.net/index.php) is used for emotions extraction. The list includes 32 word categories tapping psychological constructs (e.g., affect, cognition, biological processes), 7 personal concern categories (e.g., work, home, leisure activities), 3 paralinguistic dimensions (assents, fillers, and nonfluencies) and others, together about 4500 different words and its stems. Joint enterprises gather the measures that are caring of the awareness of technologies and goals in a community. Affordance puts constraints on the type of processes that a community member may perform within a given medium [18]. Such conditions are shaped outside the control of members as a result of long historical developments. Nevertheless, the further success of media environment depends on perceptions of affording actions [32] of the members. The members participate in the JE evolving in the community. Awareness is the boolean property that makes members of the network to be aware or not aware about network changes [12]. This property plays a central role because users feel themselves concerned or unconcerned by what they do and what is happening around them. Media centric theory of learning was born combining social learning processes and knowledge creation and is based on media operations [14,21,25]. Shared repertoire analyses mails content with different measures where the topic/term identification is the most significant. Sentence model purposes to find the sequence of words that denotes the logical sentence. The model applied in the experiment is based on sentence boundary indexes (stop lists, impossible penultimates and impossible sentence starts). The Part-Of-Speech (POS) tagging of the texts can simplify the Gist extraction [27] as it is incredibly useful for overlapping text topics that have the same sense but are named with different words. The approach of our tagger is based on Hidden Markov Models (HMM) and its implementation is done with the help of the Viterbi algorithm [35]. As a result each item of the input text is labeled with POS. 4 Metadata extraction The exploration of ME measures based on the SNA approach reveals the following characteristics for threads and members of mailing lists. – One-mail-threads are threads with one mail only. These can include cases when administrators send some informative emails (posts) which don’t initiate a discussion. The other possibility are threads that introduce some unknown, uninteresting topics to a community. If more than a half of community threads are one-mail threads, such 8 – – – – – – – a community is ineffective and non-interactive. Surely, one-mail-threads can be spam threads, though some of the examined mailing lists are observed by administrators that avoid spam posts. Threads with monologues are varieties of the one-mail-threads. A thread where its initiator posts more than one mail and gets no answer in the thread is called a monologue-thread. For instance, with the help of the monologue thread the sender is trying to explain a topic. Nevertheless, the monologue thread has no impact on community interactions. Threads without monologues positively characterize a community. In a combination with a nearly absence of one-mail-threads it can be inferred that at least ME dimensions are partly supported for the community. There are a number of different structures that can be defined with the non-monologues threads: dyads, triads, multiple dialogues, sequent conversations, balanced communications, communicative communications, etc. Unluckily initiators are users that initiate one-mail-threads and monologue threads. It depends on the network which thresholds to use to define the unluckily initiators though it can be found that the initiators receive answers in other threads, i.e. they are not so unluckily. Reply senders or answering persons are senders who reply on posts from other senders. Depending on thresholds a group of users that send replies can be defined . However belonging to the group doesn’t mean that its members are experts. They can be spammers, trolls as well as newbies in a discussion topic. Reply receivers are users that initiate threads and receive answers. A number of answers shows the importance of a discussion topic for community users. Communicators are reply senders and reply receivers. One can put the threshold for the number of replies the communicators send as well as the threshold for the number of replies communicators receive. As well it is useful to mention if the examined user is in the group of unluckily initiators. Cross-monospeakers , cross-reply receivers , cross-reply senders , cross-communicators are users that appear as monospeakers, reply receivers, etc. in more than one mailing list community. Using linguistics methods we analyze the refined content of the examined E-Learning communities. We calculate the frequencies of LIWC categories of words that appear in the threads of the mailing lists. The frequencies are stored in the database. 5 Self-classification A community possesses a set of parameters. It consists of structural parameters presented in a previous section and semantic parameters. Some categories of words with examples that were used as a locus for semantic parameters are presented in Table 1. The communities are presented as vectors where variables are their parameters. Following our purpose to find similar communities we applied hierarchical clustering (HC) and factor analysis (FA) to our data set. We utilized those two methods to find communities corresponding to the E-Learning models. With the help of the HC we found the vectors which correlate significantly. The vectors form a cluster. In its term, the FA defines a factor, i.e. the 9 Word category Words number FRIENDS 36 ANGER 364 INSIGHT 193 (understanding) FILLER POSEMO NEGEMO DISCREP (discrepancy) HOME HUMANS NONFLU (nonfluency) CERTAIN PERCEPT (observation) Included symbols or words companion, friend, mate, roomier, etc. danger, defense, enemy, rude, victim, etc. become, believe, feel, inform, seem, think, etc. 8 405 495 75 yakno, ohwell, etc. agree, casual, improve, support, sweet, etc. broke, fury, panic, temper, etc. hopeful, must, ideal, prefer, problem, etc. 92 60 7 bad, garage, mailbox, rug, vacuum, etc. adult, boy, citizen, individual, person, etc. hmm, um, etc. 82 371 clear, correct, forever, indeed, total, truly, etc. black, circle, sand, skin, etc. Table 1. Community semantic parameters consistent group of vectors. Clusters and factors reveal similar communities. As it was earlier noticed we differentiate between question-answer, innovative and disputative communities. One of the extracted clusters is similar to the question-answer E-Learning community. The number of dyadic and sequent communications is very high within the cluster. It infers that there are a lot of threads where one member asks and the other answers. Also there are some threads where two members discuss something and send each other several mails. The number of questions signs within the cluster and ”insight” and ”discrep” parameters of the cluster are significant thus these indicate the query-explanation nature of the texts within the cluster. The other cluster looks like a very interactive system with a lot of users characterized as the ”reply sender” . The semantic parameters, e.g. explanations , disagreements and quarrels, infer presence of discussions in the cluster. We assume that the cluster is a realization of the disputative E-learning community. Considering groups of communities defined by FA, one of the group has two options: the disputative community or the question-answer community. The identity of the question-answer community are ”dyadic” communications that are not the case for the group, although it has a lot of semantic parameters which fit to the question-answer community. For the disputative community high weights of ”reply sender” and ”reply receiver” parameters within the group are important, though the ”reply sender” parameter is only significant for the group. The communities in the group might be a kind of mixture between two options. The discussions appear not only between two members and are performed using words that denote doubts and disagreements . 10 The other group we defined with the help of the FA is positioned between the disputative and the innovation communities. There are ”communicator” (refers to innovation community), ”reply receiver” and ”reply sender” (two last refer to disputative community) parameters presented. After comparing the results of the HC and the FA approaches we concluded that the HC focuses on the whole set of variables that represents a community. The FA concentrates on the variables that even after normalization are bigger than the others. For our data set the difference of the approaches means that the HC pays more attention to structural parameters and organizes communities according to them while the FA focuses on the semantic parameters. 6 Conclusion and outlook The focus of the paper was the E-Learning community, its structures and content. The theory that we used to study the community notion is CoP [37]. According to CoP dimensions, a community is a CoP if there are interactions of members within the community (ME); the community possesses technologies, rules and policies (JE); the community members are aware of community knowledge (SR). The model of CoP for E-Learning communities was illustrated with the help of the i* modelling. Each actor of a community is presented with the same prototype and is completed by adding a set of parameters that identifies the actor. Using the ANT [28] model it makes the definition of relations between actors easier and consequently the classification of them easier. Before applying the analysis we refined the data. The result of the refinement is a 10% decrease in the number of threads of mailing lists and consistency of the mailing lists content. The paper demonstrates how E-Learning communities can observe themselves and basing on the observations how they can model themselves. The work we have done opens several issues that can possibly improve the results we achieved and defines new topics. – It is useful to analyze E-learning communities according to processes that can be performed within media and how these processes influence the learning process? – Can the i* models of E-learning communities can be effectively extended by considering semantic parameters? – Is the application of linguistic techniques important and will it open many possibilities for the semantic analysis of E-learning communities? – Which analysis is appropriate for determining all components of CoP and consistent groups of communities? There will be even more transitions between informal and formal learning phases in future. Learners will spend some of their lifetime still in institutions offering formal education, training and qualification. However, life-long learning and continuing education create also the need for more informal learning situations like learning in communities which share a common practice. In the roadmaps created by major European research intitiatives theses needs are clearly expressed. However, the long tail of learning puts complex decision processes on the learners. Our research investigates these decision processes by the means of applied game theory. We will find models to help learning communities finding optimal strategies in selecting learning materials and the right ratio between experts and novices in the communities. 11 Acknowledgements This work was supported by the German National Science Foundation (DFG) within the collaborative research centers SFB/FK 427 “Media and Cultural Communication” and within the research cluster established under the excellence initiative of the German government “Ultra High-Speed Mobile Information and Communication (UMIC)”. We thank our colleagues D. Denev, M. Spaniol and H. Nan for the inspiring discussions. References 1. D. Albrecht, R. Arnold, W. Bauerfeld, A. Bode, E.-M. Bruch, U. Cress, J. Haake, R. Keil, J. Kuper, W. Nejdl, G. Reinmann, C. Rensing, J. Schaper, and V. Zimmermann. Web 2.0: Strategievorschläge zur Stärkung von Bildung und Innovation in Deutschland. Technical report, Bundesministerium für Bildung und Forschung (BMBF), 2007. 2. C. Anderson. The Long Tail: Why the Future of Business is Selling Less of More. Hyperion:New York, 2006. 3. A.-L. Barabási. The Physics of the Web. Physics World, 14:33–38, 2001. 4. A.-L. Barabási. Linked: The New Science of Networks. J. Artificial Societies and Social Simulation, 6(2), 2003. 5. S. Baron, J. Field, and T. Schuller. Social Captial - Critical Perspectives. Oxford University Press, Oxford, 2000. 6. J. Seely Brown and R.P. Adler. Minds on Fire: Open Education, the Long Tail, and Learning 2.0. EDUCAUSE Review, 43(1):16 – 32, 2008. 7. K. Christensen, D. Levinson, R. Jarrett, D. R. Judd, W. Metcalf, R. Moudry, R. Oldenburg, S. Salamon, T. Sander, M. Shuman, B. Wellman, and M. Zuckerman, editors. Encyclopedia of Community: From the Village to the Virtual World. New York: Sage, 2003. page xxxi. 8. D. Cohen and L. Prusak. In Good Company - How Social Capital Makes Organizations Work. Harvard Business School Press, Boston, MA, 2001. 9. N. Couldry. Actor Network Theory and Media: Do They Connect and On What Terms? Chapter for A. Hepp et al. (eds) Cultures of Connectivity, London: School of Economics and Political Science, pages 1-14, 2004. 10. A. Degenne and M. Forsé. Introducing Social Networks. SAGE Publications, London, 1999. 11. R. Diestel. Graph Theory, volume 173 of Graduate Texts in Mathematics. Springer-Verlag, New York, 2nd edition, 2000. 12. P. Dourish and V. Bellotti. Awareness and Coordination in Shared Workspaces. In CSCW ’92: Proceedings of the 1992 ACM Conference on Computer-Supported Cooperative Work, pages 107–114, New York, NY, USA, 1992. ACM Press. 13. G. Fischer, M. Rohde, and V. Wulf. Community-based Learning: The Core Competency of Residential, Research-based Universities. International Journal of Computer-Supported Collaborative Learning, 2(1):9–40, March 2007. 14. J. Fohrmann and E. Schüttpelz. Die Kommunikation der Medien. Niemeyer, Tübingen, 2004. 15. L. Freeman. Centrality in Social Networks. Conceptual Clarification. Social Networks, 1:215–239, 1979. 16. Bundesministerium für Bildung und Forschung. Neue Medien in der Beruflichen Bildung. Technical report, Bundesministerium für Bildung und Forschung (BMBF), 2007. 17. G. Geser. Open Educational Practices and Resources - the OLCOS Roadmap 2012. Technical report, Salzburg Research, 2007. 18. J. J. Gibson. The Ecological Approach to Visual Perception. Hillsdale, NJ: Laurence Erlbaum Associates, Inc, 1979. 19. R. A. Hanneman. Introduction to Social Network Methods. University of California, 2001. On-line text book, http://www.faculty.ucr.edu/ hanneman/nettext/, last access on 12.03.07. 12 20. C. Haythornthwaite and A. Gruzd. A Noun Phrase Analysis Tool for Mining Online Community. In In Proceedings C & T’07, Michigan, USA, pages 67–86, 2007. 21. L. Jäger and G. Stanitzek, editors. Transkribieren - Medien/Lektüre. Wilhelm Fink Verlag, Munich, 2002. 22. V. Kamtsiou, A. Naeve, M. Kravcik, D. Burgos, V. Zimmermann, R. Klamma, M.A. Chatti, P. Lefrere, J. Dang, and T. Koskinen. A Roadmap for Technology Enhanced Learning (TEPL). Technical report, PROLEARN Network of Excellence, 2008. 23. R. Klamma, Y. Cao, and M. Spaniol. Watching the Blogosphere: Knowledge Sharing in the Web 2.0. In N. Nicolov, N. Glance, E. Adar, M. Hurst, M Liberman, J. H. Martin, and F. Salvetti, editors, International Conference on Weblogs and Social Media, Boulder, Colorado, USA, pages 105–112., March 26-28, 2007. 24. R. Klamma, M. Spaniol, Y. Cao, and M. Jarke. Pattern-Based Cross Media Social Network Analysis for Technology Enhanced Learning in Europe. In First European Conference on Technology Enhanced Leanring (EC-TEL 2006), pages 242–256, 2006. 25. R. Klamma, M. Spaniol, and M. Jarke. "Do You Know a Similar Project I Can Learn From?" Self-Monitoring of Communities of Practice in the Cultural Sciences. In ICITA ’05: Proceedings of the Third International Conference on Information Technology and Applications (ICITA’05) Volume 2, pages 608–613, Washington, DC, USA, 2005. IEEE Computer Society. 26. J. Kleinberg. Authoritative Sources in a Hyperlinked Environment. J. ACM, 49:604–632, 1999. 27. G. Lakoff and M. Johnson. Metaphors We Live By. Chicago: University of Chicago Press, 1980. 28. J. Law. Notes in the Theory of the Actor Network: Ordering, Strategy and Heterogeneity. System Practice, 5:379–393, 1992. 29. V. I. Levenshtein. Soviet Physics Doklady 10, chapter Binary codes capable of correcting deletions, insertions, and reversals, page 707–710. 1966. 30. P. Mika. Social Networks and the Semantic Web. Springer Science + Business Media, LLC, 2007. ISBN-13: 978-0-387-71000-6. 31. G. Mishne and N. Glance. Leave a Reply: An Analysis of Weblog Comments. WWW2006, 3rd Annual Workshop on the Weblogging Ecosystem: Aggregation,Analysis and Dynamics, Edinburgh, 23-26 May, 2006. Retrieved from http://staff.science.uva.nl/ gilad/pubs/www2006blogcomments.pdf, last access on 25.06.07. 32. E. D. Mynatt, V. L. O’Day, A. Adler, and M. Ito. Network Communities: Something Old, Something New, Something Borrowed... Computer Supported Cooperative Work: The Journal of Collaborative Computing, 7:123–156, 1997. 33. M. Spaniol, R. Klamma, and M. Jarke. Die Listen der Evidenz, chapter Selbstbeobachtungswerkzeuge zur cross-medialen, sozialen Netzwerkanalyse kulturwissenschaftlicher Communities, pages 239 – 254. ISBN 3-8321-7979-8. DuMont, Köln, 2006. 34. F. Vega-Redondo. Complex Social Networks. Cambridge University Press, 2007. 35. A. J. Viterbi. Error Bounds for Convolutional Codes and an Asymptotically Optimum Decoding Algorithm. IEEE Transactions on Information Theory, 13(2):260–269, April 1967. The Viterbi decoding algorithm is described in section IV. 36. S. Wasserman and K. Faust. Social Network Analysis. Cambridge University Press, 1994. 37. E. Wenger. Communities of Practice. Learning, Meaning and Identity, chapter Communities, pages 72–85. Cambridge University Press, Cambridge, UK, 1998. 38. E. Yu and J. Mylopoulos. Towards Modelling Strategic Actor Relationships for Information Systems Development – With Examples from Business Process Reengineering. In Proceedings of the 4th Workshop on Information Technologies and Systems, 1994.