The Internet, Policy & Politics Conferences

Oxford Internet Institute, University of Oxford

Francesco Bailo: Mapping online political talks through network analysis: A case study of the website of Italy's Five Star Movement

Francesco Bailo, The University of Sydney


In this paper I apply network analysis and data mining to map in detail the relations among the 84,203 users and 548,240 postings, published between 2009 and 2014, of the Internet forum of Italy's Five Star MoVement (M5S) to unravel characteristics of online political talks.

This paper aims at improving the scholarly understanding of the argumentative political discourse but has also applications for the enhancement of computer-mediated public reasoning.

* Framework

Network analysis has two key attributes that make it extremely effective in querying online content: it leverages the linked structure of the Internet data and it scales. But network analysis is not an interpretative tool; it maps edges and nodes without qualifying the relations nor the meaning of their position within the network. That is, in the case of online political talks, network analysis cannot provide by itself an answer to two fundamental questions: what is being talked and who is talking. 

The theory framing my reading of the network analysis output is based on two concepts that respectively provide a useful interpretative dictionary to help answering the What and Who questions: the concept of 'everyday political talk' [2] and the concept of 'networked self' [5].

Acknowledging the concept of 'public sphere' [3], I assume the textual exchanges between online users to be a form of 'dialogic deliberation' [4] that takes place within a 'weak public sphere' [1] in charge of shaping public opinions. The informality of the political talk - 'not always self-conscious, reflective, or considered' [2, 211] and 'produced by nonpurposive, nonstrategic, nonsuccess-oriented social interactions' [4, 54] - does not undermine its deliberative value. Everyday political talks are indeed part of a broader 'deliberative system' [2], which also includes a 'strong public sphere' [1] with its 'instrumental' [4] and binding forms of deliberation.

I conceive the user as a 'networked self' who is simultaneously part of 'social networks that are more diverse and less overlapping than [...] previous groups' [5, 9]; and who does not dissociate in distinct online and offline selves but instead 'a single self that gets reconfigured in different situations' [5, 126]. 

In other words, my assumption is that the political conversation that occurs online - out of formal deliberative rules and without the goal of producing binding decisions - has political importance and deliberative value because its participants are in effect public actors contributing to public reasoning. 

* Hypotheses 

Through the analysis of the characteristics and relations among users and discussion threads I test five hypotheses. Specifically the first two will test my theoretical framework - can online forums be a fair representation of the broader national polity and public debate? - while the last three will explore participant behaviours. 

H1: There is a correlation between public events and online participation;
H2: Online political participation decreases gender imbalances;
H3: Users engage in discussions on limited topics;
H4: Online political participation fragments in conversations among like-minded;
H5: Forum activity shows a property of preferential attachment with a 'the-rich-get-richer' effect. 

* Case selection

Online communities generally demonstrate too high a level of homophily among users to be illustrative. But I argue that online political talks do not necessarily prevent diversity of opinions or topics debated. The discussion forum the M5S provides an interesting case study because of the diverse political orientations of the participants - the M5S has been attracting voters (25.5 percent of the popular vote in 2013) from the entire political spectrum - and the wide range of discussion topics.

Within the forum I then select two issues, immigration policy and coalition government support, and I map overtime the interactions between participants and threads of different - and very different - political orientations. 

* Methodology

After indexing the online content through Google Custom Search and Disqus API, I scraped 86,943 threads and 461,297 comments published between February 2009 and January 2014. The data was cleansed by removing spam comments, duplicated threads and integrating two different user sets (notably opening a new thread and commenting an existing one require two different user identifications).

A two-mode network was initially drawn on the data with two kinds of nodes: users and threads. The network was then projected into two one-mode networks; a user network, with edges representing discussions that users have in common, and a discussion network, with edges representing users that threads have in common. 

For the debate analysis on two issues, I identified with statistical text mining techniques the threads of interest, I extracted them to form two single-issue discussion networks, and I coded each thread according to the political orientation of the opening post. 

* Findings

Results do not suggest that talks on the forum might be not representative of what is talked in the broader society. The assumptions of H1 are supported by the data indicating that forum activity is in sync with all major political events such as elections. Conversely the quality of participation in terms of gender balance (H2) does not show any significant difference from societal trends: in the forum 12 percent of users have been identified as women, in line with a very low (21 percent) presence of women in elected offices across the country. 

H3 is confuted by the data, suggesting that probably motivations to participate are not limited to special interests but might also involve sense of community and empowerment. Similarly H4 is not supported by any finding: polarisation and fragmentation - even though differences do show up - do not appear to be a feature of the discussion. Not surprisingly H5 is strongly supported by the data: few users are responsible for most of the comments, and few threads attract most of the comments. 


[1] Fraser, N. (1990). Rethinking the Public Sphere. Social Text, (25/26), 56–80.

[2] Mansbridge, J. (1999). Everyday Talk in the Deliberative System. In S. Macedo (Ed.), Deliberative politics: essays on democracy and disagreement. New York, NY: Oxford University Press.

[3] Habermas, J. (1989). The structural transformation of the public sphere (T. Burger & F. Lawrence, Trans.). Cambridge, MA: MIT Press. (Original work published 1962)

[4] Kim, J. & Kim, E. J. (2008, January). Theorizing Dialogic Deliberation. Communication Theory, 18(1), 51–70

[5] Rainie, L., & Wellman, B. (2012). Networked. Cambridge, MA: MIT Press.

Francesco Bailo