Scalability and the Social Dynamics of Communication. On Comparing Social Network Analysis and Communication-Oriented Modelling as Models of Communication Networks (with S. Albrecht, M. Lübcke, C. Schlieder)

Please download to get full document.

View again

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
 7
 
 

Career

  Internet communication is a major challenge for anyone claiming to design scalable multiagent systems. Millions of messages are passed every day, referring to one another and thus shaping a gigantic network of communication. In this paper, we compare
Share
Transcript
  Scalability and the Social Dynamics of Communication.On Comparing Social Network Analysis andCommunication-Oriented Modelling as Models of Communication Networks Steffen Albrecht 1 , Maren L¨ubcke 1 , Thomas Malsch 1 , and Christoph Schlieder 2 1 Hamburg University of TechnologyDepartment of Technology AssessmentSchwarzenbergstr. 95, 21071 Hamburg, Germany { steffen.albrecht, maren.luebcke, malsch } @tu-harburg.de 2 Bamberg UniversityDepartment of Applied Computer ScienceFeldkirchenstr. 21, 96045 Bamberg, Germany christoph.schlieder@wiai.uni-bamberg.de Abstract.  Internet communication is a major challenge for anyone claiming todesign scalable multiagent systems. Millions of messages are passed every day,referring to one another and thus shaping a gigantic network of communication.In this paper, we compare and discuss two different approaches to modelling andanalysing such large-scale networks of communication: Social Network Anal-ysis (SNA) and Communication-Oriented Modelling (COM). We demonstratethat, with regard to scalability, COM offers striking advantages over SNA. Basedon this comparison, we identify mechanisms that foster scalability in a broadersense, comprising issues of downscaling as well. 1 Introduction Internet communication is a very large-scale process with millions of messages passedevery day. Uncountable emails are sent and received, websites are visited by growingnumbers of users in search for information, and masses of contributions are publishedin Internet forums such as Usenet discussion groups. With figures like, for instance, the700.000.000 messages in the Usenet archives on Google’s servers, 1 increasing by ca.150.000.000 messages every year, Internet communication appears to be an ideal casefor approaching issues of scalability in multiagent systems (MAS).In viewing communication as a central element to consider in the design of MAS,we follow the methodology of Communication-Oriented Modelling (COM) that wasdevelopedby Malsch and Schlieder[1]. COM is supposedto complementand reinforceagent-oriented modelling (AOM), today the standard approach to distributed artificialintelligence (DAI). In contrast to AOM, COM models communication as a stream of messages connected to one another by references. It does not conceive of communica-tion as a transmission of information from sender to receiver, as it is typical for agent 1 See groups.google.com. K. Fischer, M. Florian, and T. Malsch (Eds.): Socionics, LNAI 3413, pp. 242–262, 2005.c  Springer-Verlag Berlin Heidelberg 2005  Scalability and the Social Dynamics of Communication 243 communication languages in current agent platforms. From the COM perspective, in-dividual agents and their internal operations are only peripheral elements. The focus ison the emergent structural properties of communication.In this paper, we want to evaluate COM’s potential for the design of scalable mul-tiagent systems. We do this by comparing COM to another way of modelling socialstructures: Social Network Analysis (SNA). Both COM and SNA can be regarded associological models, derived from the analysis of social phenomena. Both model com-municationin the form of networks,thus allowing to analyse and simulate variouscom-municativephenomenain MAS as well as humansociety. However,while SNA followsan actor-centric approach and is largely static, COM puts communication in the centreof analysis and focuses on its dynamics.Our hypothesis is that COM is better suited to model communication on a largescale. By systematically exploiting the temporal dimension of dynamic network repro-duction, it should be capable of accelerating and up-scaling its message turnover to anextent that meets the demands of real world mass communication in the Internet. Inthe following section we demonstrate that the small differences between both modelsactually have a major impact on our ability not only to describe and explain, but alsoto simulate and empirically analyse large-scale processes of communication in the In-ternet. Such a potential is also relevant for future perspectives—once a powerful toolhas been developed, it can be used to support human beings as well as artificial agentsengaged in such processes of communication,and an important step towards the designof large-scale open systems in the sense of Hewitt [2] would be made.As we think of such practical relevance for the design of MAS, we have to recon-sider the notion of scalability. Large numbers,as suggested by much of the literature onscalability, is not the only important aspect. Scaling up is just one option for an MAS tocope with a changing demand. The underlying issue, as we see it, is the ability to reactflexiblyto changingenvironments.Scalingdown,then,canbeimportantas well, e.g.,toconcentrateon solving difficult and complex problems.In Section three, we present ournew perspective and discuss different strategies to cope with scalability in this sense.The final section concludes with a reflection on SNA and COM’s ability to modelscalable communication networks. It summarizes the results of our comparison andhighlights the advantages of the COM model. However, as a field of intensive develop-ment, new work in SNA proposes to have interesting implications. Cross-fertilizationbetween COM and SNA will be a promising option for future work. 2 SNA and COM as Models for Scalable Communication Systems One of sociology’s central purposes is the construction of models of the social world.Models serve as a link between abstract theories and empirical reality [3]. They trans-late the social theoretic perspective into a description of the relations between relevantentities, each of which can be tested by observation. In its long tradition, sociologyhas developed a large number of models at various levels of abstraction and complex-ity. Some of the sociological models are directly related to specific domains of socialphenomena, others claim to be relevant for a broader range of the social world. Due totheir generality, the latter can be regarded as methodological tools for describing andanalysing social phenomena.  244 Steffen Albrecht et al. The transfer of concepts and models from sociology to DAI and to the develop-ment of MAS is one of the acclaimed outcomes of Socionics [4]. Along the lines of the so-called “computational reference”, we examine how sociological models of com-munication can be used as models for the construction of large-scale communicationprocesses in MAS. While research in DAI has acknowledged the need to move fromsingle messages to sequences of message exchange, i.e., ‘conversations’ [5], the mod-els we are interested in go one step further, trying to enhance our understanding of communication on the level of agent societies that comprise vast numbers of agents.Such an understandingis the first step towards—and a necessary prerequisiteof—usingcommunication structures in the design of MAS.For this purpose, we have identified two sociological models that seem worth acloserinvestigation:SocialNetworkAnalysisandCommunication-OrientedModelling.Both are rather abstract models, but nevertheless providemeans that are precise enoughto conceptualize communication processes. SNA has a long tradition of research, but isless specific than COM—a verynew approach,outlined only recently.Despite apparentsimilarities of their visualizations, there is a considerable degree of difference betweenboth. We will present both models and compare them to examine the impact of theirdifferences on their ability to serve as models for scalable communication systems. 2.1 Social Network Analysis: A Guide to Modelling Communication? Within the social sciences, interest in using Social Network Analysis to describe andanalyse social structures has been steadily increasing. In the early days of ‘sociomet-rics’ in the 1930s, researchers started to use graph drawings to describe interpersonalstructures. This was the birth of the “network perspective” in the social sciences. In thefollowing decades and until today, more and more social phenomena were describedwith the help of network models: Studies using SNA range from small group researchto research into organisational structures as well as the analysis of, e.g., internationaleconomictransactions.The results ofthese studieshaveprovedtheviabilityofapplyingthe network perspective in sociology, so that today, network analysis is seen as “one of the most promising currents in sociological research” [6].What is it that makes SNA so attractive, and what makes it seem an interestingcandidate for modelling large-scale processes of communication? First, a network is avery flexible model for various forms of structural patterns. SNA presupposes only twobasic entities, nodes and edges. It offers a number of mathematical methods to derivemeaningful information out of the various combinations that can be observed or mod-elled with these two entities. Second, concerningthe analysis of empirically observablenetworks, SNA offers methods that are highly sophisticated from a mathematical pointof view. They build on algorithms forming the state of the art in statistics (e.g., hierar-chical clustering, factor analysis) as well as special procedures based directly on graphtheory (e.g., triad census) (cf. [7], [8]). However, SNA should not be seen as a mere toolbox for the quantitative analysis of structural data. This would ignore a vast array of literature trying to make sense of thenetwork phenomenon in theoretical terms. Thus, thirdly - as Wasserman and Faust putit in their seminal textbook on SNA - , “network analysis, rather than being an unre-lated collection of methods, is groundedin important social phenomena and theoretical  Scalability and the Social Dynamics of Communication 245 concepts.” [7] That is, there is a body of theoretical concepts and hypotheses laying thefoundations for structural analysis, and both theory and methods are intimately inter-woven with one another.InthistraditionofSNAasaresearchperspective(ratherthanameresetofmethods),social network models largely concern sets of actors and their relations. Although inprinciple, the graph theoretic algorithms of SNA can be applied to whatever basic unitone is interested in, SNA is typically actor-centric, focusing on individual or collectiveactors as the nodes of the network . 2 Thus, communication networks are conceptualizedas exchange networks with a set of actors engaging in the exchange of messages. 3 There are good reasons to attempt to model processes of communication with thehelp of SNA. All theories conceiveof communicationas an intrinsically  relational  phe-nomenon. Communication establishes ties between actors in turning from utterance toreception and vice versa. SNA seems to be well suited to model communicationsince ithas gained respect particularly for capturing the relational aspects of social structures.Despite being based on such simple constructs as nodes and edges, it has proven to beable to grasp even highly complex structures resulting from the selective combinationof these basic elements.Applying SNA to model especially  large-scale  communication processes is a lessevident choice. The research methods of SNA srcinated in small group research, andstill today many algorithms are based on matrix representation that pose difficultieswhen growing to large scales [14]. However, recent applications of network measuresto the graph of the world wide web [15] show that in general it is possible to model andanalyse even very large networks. It depends on the methods that need to be applied tothe networks whether scale is a problemor not. We will examinethis issue more deeplyafter we have introduced a second way of modelling communication processes: COM. 2 This does not mean that SNA is employing an overly simplistic concept of agency. Actors canbe human beings, but also collective actors like families (as in Padgett and Ansell’s study onthe marriage strategies of the Medici and their peers [9]), organisations, or even nation states(as in a study of the international telephone network [10] and of the Eurovision song contest[11]). But in any of these cases, actors are relatively stable social entities with a certain degreeof autonomy in their behaviour. Furthermore, while we do identify SNA with an actor-centricperspective, this does not mean that within the broad community interested in SNA there areno other applications of the network perspective. Examples include information networks (asin citation analysis) or semantic networks. As extensions to the actor-centric perspective of SNA, these are discussed in the final section of this paper. 3 A typical—even canonical—study of communication networks is Freeman’s analysis of anearly computer conferencing system [12]. He studies a network of 50 scientists using a CMCsystem, and measures their relations based on awareness, acquaintanceship and exchange of messages. The structural properties of the network are analysed by looking at the amountof messages exchanged between each pair of actors in a given time. Thus, the scientists andtheir relational properties are in the centre of analysis, and the author even claims that “in onesense, then, the study of the sociology of science is the study of links among persons.” [12]For a similar application of SNA to communication networks, see [13].  246 Steffen Albrecht et al. 2.2 Communication-Oriented Modelling In contrast to SNA, COM is specifically designed to understand and analyse the com-plexity and temporality of social communication. It is based on a social theory of com-munication and provides conceptual means focused on modelling communication pro-cesses. COM was proposed as a model for multiagent communication by Malsch andSchlieder [1]. Here, we summarize the theoretical foundations only briefly and concen-trate our discussion on issues of large scale. Theoretical Foundations.  As we have noted above,communicationin COM is not or-ganised alongthe lines of agent-to-agentrelations as in AOM. In communicationon thelevel of society, as for instance on the Internet, we can observe patterns of communica-tionorganisedasmessage-to-messagerelations.Thismeansthatmessages refertoothermessages in an ongoing process of weaving and reweaving complex webs of commu-nication. Moreover, messages usually are not sent to a specific receiver, but published“to whom it may concern”. Thus, whenever a message is published for an audiencerather than sent to a receiver, and whenever communication is dominated by messagesreferringto other messages rather than agents influencingother agents’ actions, it is notthe agent but the communication that should be considered as the foundational unit of analysis.Communication consists of two types of operations: reception (understanding amessage) and inception (producing a message). Inception and reception are definedas the temporal elements (or elementary operations)of social communication.They arecomplementary operations. Messages cannot be connected with each other by either areception or an inception alone. Both operations must be activated and carried out in atemporal order, i.e., a predecessor message must have been received before any succes-sor message can be inceived etc. Defining inception and reception as the ope elementsof social communicationmeans to compare them to the elements in biological systems,i.e., biologicalcells in a livingbody.These are permanentlyreproducedandexchanged,and so are communicative operations. Hence, and this is in accordance with what maybe called the communicative turn in sociology (cf. [16], [17]), reception and inception are construed as the temporal “stuff” that communication networks are made up of.In contrast to the transient character of communicative operations, messages canbe relatively persistent. Messages are empirical sign-objects and—again in contrast tocommunicativeoperations—beingempirical,theycanbeobserved.InlinewithPeirce’ssemiotics and Mead’s concept of symbolic interactions, a message is a perceivable,em-pirically observable object [18]. It is a meaningful object, or in Mead’s terminology,a significant object. Being meaningful and empirically observable, messages point outto communicative operations, which, in turn, are unobservable. Whenever a messagerefers to another message, we can reasonably assume that the preceding message hasbeen received, and that this reception has triggered the inception of the referring mes-sage.Methodologically,wecannotobservecommunicationatthelevelofits elementaryoperations.Thus, to draw inferencesabout the inceptionsand receptionsactually takingplace, we have to observe the pattern of referencing.In COM, messages are activated, deactivated,or reactivated in a continuousprocessof selective referencing. There are always messages that are drawn on again and again
Related Search
Similar documents
View more
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x