Dynamics of online discussions about politics: a function of structural network properties, mass media attention or emotional utterances? Dirk Oegema VU University Amsterdam d.oegema@fsw.vu.nl Shenghui Wang VU University Amsterdam s.wang@fsw.vu.nl Jan Kleinnijenhuis VU University Amsterdam J.Kleinnijenhuis@fsw.vu.nl ABSTRACT Easily accessible online discussion forums have greatly low- ered the hurdles for citizens to join political debates. In this paper we employ automated content analysis of online discussions which offers as a new opportunity to assess the agenda of online discussions unobtrusively. We focus on the twofold Research Question how previous communication content and previous network structure affect (1) the current communication content and (2) the current network struc- ture. By analysing the dynamics of the online discussions together with the related political content and emotional utterances, we obtain a new perspective towards the con- nection of online behaviour in the web and the traditional political science research. Keywords Social network analysis, content analysis, political agenda setting, news effects 1. INTRODUCTION Unlike the respondents in survey research, the partici- pants in online discussions are not a random sample of the population. At best they are a representative sample of the involved citizens who are able and willing to spend time to share information and to express their opinions. They resemble the Aktive ¨ Offentlichkeit (active citizenry) as op- posed to the Passive ¨ Offentlichkeit (passive citizenry) in the terms of the German political philosopher Ralf Dahrendorf. He argued that opinions of the active citizenry are repre- sentative in a political sense for the opinions of the passive citizenry, as long as the hurdles for passive citizens to be- come active are sufficiently low [4]. Online discussion fo- rums have lowered the hurdles for participation in political debates, since everybody with access to the Internet is free to participate. Unlike answers to survey questions, which are often unimportant from the perspective of citizens, the postings to online discussions are really important, salient and top of mind from the perspective of citizens. In this paper two interrelated research questions with re- spect to political online discussions will be addressed. The first question is what drives the development of topics and opinions in online discussions. Here we will start from agenda setting research [8]. Agenda setting theory states that citi- Copyright is held by the authors. Web Science Conf. 2010, April 26-27, 2010, Raleigh, NC, USA. . zens will take over the themes and topics addressed by the mass media, although they will presumably not believe in the solutions suggested in the mass media. The media may not determine what people think, but they are important because they may determine where people think about. It is to be expected that participants in online discussions will respond to themes addressed in the mass media. Agenda setting cannot be the only answer, however. Respondents respond primarily to each other. It is expected that they will respond especially to participants who have gained a prominent position in the online network. Thus, we will try to explain the development of topics and opinions also from the development of the social network of the participants. The second question to be addressed is therefore what drives the development of the social network of the partici- pants in online discussions. These networks typically show a “long tail distribution,” with a relatively small core of heav- ily interacting participants and a large periphery of persons who connect only a few times to others, preferably with al- ready connected participants. In [1], Barab´ asi and Albert explained that growth and preferential attachment – of new members getting connected with the ones in their neigh- bourhood who are already heavily connected – is also the mechanism to explain the varying popularity of web pages. Growth and preferential attachment gives rise to a nega- tive exponential distribution (known as Zipf’s law in lin- guistics). For example, the probability that k documents are hyperlinked to a given webpage can be written sim- ply as a negative exponential distribution with parameter λ =2.1: P (k) ∽ k -2.1 . Preferential attachment cannot be the only answer for the popularity (=indegree centrality) of discussants in online discussions, however. In addition to the previous state of the social network – as measured by network properties such as popularity (indegree centrality) and brokerage (betweenness centrality) of network vertices – actual communication content may play a significant role. Who was outrageous about which political actors? Who remained quiet and who used a language of hatred and dis- gust? On the basis of the theory of agenda setting one would expect that it will also matter who follows the agenda of the mass media. We summarise the twofold research question as follows: how will previous communication content and previous net- work structure affect (1) the current communication content and (2) the current network structure? In this paper we will employ automated content analy- sis of online discussions which offers a new opportunity to assess the agenda of online discussions unobtrusively. This 1