Main-01.png Read-01.png Recipes-01.png Tactics-01.png Cases-01.png Tools-01.png

Detecting the Bias of Issues with Twitter

On Bias Detection as an Online Method
While certainly a happening issue in journalism and media studies, the issue of bias has not received that much attention in sociology recently, mainly for historical reasons we believe. In recent decades, sociologists have radically extended the notion of bias and partisanship: they have sought to show that all knowledge is informed by social and political interests and contexts (Bloor, 1982). As a consequence, the detection of bias in individual cases became less urgent and relevant. However, in the context of the digitization of knowledge culture, methods for bias detection are making a comeback. Digitization has put the issue of the partisanship of knowledge back at the forefront of public debate: there are significant concerns about the bias and partisanship implicit in the selection and disclosure of information online. To give one indication of this: the new media critic Evgeny Morozov (2012) has recently called for a return to a prohibitive regulation of ‘psuedo-science’, proposing that Google should introduce a red flag system highlighting sources that are not scientifically accurate, for instance anti-vaccine organisations.

In order to address the politicization of information, computational methods are currently being developed for analysing the bias and leaning of networked information and knowledge online. Computer scientists have formulated methods for the polarization of data sets that make it possible to determine the leaning of information sources and key words (Borra and Weber, 2012; Weber et al., 2012a; Weber et al., 2012b). Weber, Borra and others have developed such polarization methods, which make it possible to detect whether given information sources or key-words are biased towards the right or the left, conservative or progressive. These methods rely on co-occurrence measures: selecting actors with an online presence and known political leaning, the approach proceeds to determine which issue terms are associated with these biased users. As they rely on such co-occurrence measures to do so, these methods are not dissimilar to the actor-issue-network maps plotted by Actor-Network Theory and related approaches (Latour, Venturini and Jensen et al, 2012;, 2000; Marres, Gerlitz et al., 2012).

While originally developed for the analysis of search engine query data (Borra and Weber, 2012; Weber et al., 2012a), the method has more recently been applied to Twitter data (Weber et al, 2012b), identifying the degree of political leaning of hashtags as related to US politics. During the upcoming workshop we will explore the possibilities of detecting the bias not just of information and actors, but of issues. The aim is to develop a more open-ended and contextual approach to bias, which treats bias as a dynamic, issue-specific property of objects and terms. To do so is to build further on analysis of bias developed in work on issue-networks on the Web and on partisanship mapping with Google (Marres and Rogers, 2008 and Rogers, Borra et al., 2008).

In the upcoming workshop seeks to explore methodological concerns, relevant visual tactics, scenarios of use and data analysis techniques related to bias detection. We will do this by working with a specific data set, namely Twitter data relating to a recent issue of Intellectual Property, and a series of methods developed out of the first issue mapping workshop.

Data set
For this workshop we will work with a specific data set, namely Twitter data relating to a recent issue of Intellectual Property, the Anti-Counterfeit Trade Agreement (ACTA). We have collected all tweets mentioning this issue (ACTA) for the period of 20 May to 21 June, 2012, resulting in a data set of 45854 tweets in total.

You can find more information about ACTA here:
Arthur Charles (2012) Acta down, but not out, as Europe votes against controversial treaty, The Guardian, Wednesday 4 July

We have polarized our ACTA Twitter data, using a method inspired by the polarization methods of Borra and Weber (2012) and Weber et al. (2012b), and elaborating it further.

To polarize our data set (divide it up in pro- and con- tweets), we have used two issue-networks related to pro and anti ACTA organisations on the Web (Marres and Rogers, 2008). Using starting points proposed to us by two issue experts, Vera Franz and Becky Hogge, we located two issue-networks using Issue Crawler (link): one pro-ACTA network and one con-ACTA network [Include Figures].

On this basis we divided our Twitter data into three camps:

1. Anti-ACTA: all tweets containing URLS present in the Anti-ACTA issue network

2. Pro- ACTA: all tweets containing URLS present in the Pro-ACTA issue network

3. Neutral-ACTA: all tweets containing URLS present in both the Anti-ACTA and Pro-Acta issue networks


Anti-ACTA network on the Web, located with the aid of IssueCrawler, June 21, 2012


Pro-ACTA network on the Web, located with the aid of IssueCrawler, June 21, 2012

Initial Analysis
We have produced an initial textual analysis of the three camps in order to guide explorations during the workshop (polarization statistics, word frequency analysis and co-word analysis per camp). These initial, raw findings provide the empirical material to be explored in the workshop [..]


Political book buying patterns on Amazon, updated every hour.

Benkler, Y. (2012) ' Truthiness and the Networked Public Sphere', Symposium on Truthiness in Digital Media, Berkman Centre for Internet and Society, Harvard University, March 6-7.

Bloor, D. (1982) 'Durkheim and Mauss Revisited: Classification and the Sociology of Knowledge', Studies in History and Philosophy of Science 13 (4): 267--97.

Borra E., Weber, I. (2012). Political insights: exploring partisanship in Web search queries. First Monday, 17(7). Available at Accompanying tool online at Political Search Trends, Yahoo! labs. (2000), The GM Food Debate, poster in Preferred Placement, R. Rogers (Eds) Maastricht: Jan Van Eyck Academie.

Latour, B, T. Venturini and P. Jensen (2012) The Whole is Always Smaller Than Its Parts’ A Digital Test of Gabriel Tarde’s Monads, British Journal of Sociology,

Marres, N. and Rogers, R. (2008) Subsuming the Ground: How local realities of the Ferghana Valley, the Narmada Dams, and the BTC pipeline are put to use on the Web. Economy and Society 37, 2: 251-281.

Marres, N., Gerlitz, C., et al., (2012) Climate Change on Twitter: Issue Lifelines. DMI Summer School Project, University of Amsterdam

Morozov, E. (2012) Warning: This Site Contains Conspiracy Theories, Slate, Jan. 23,

Rogers, R. Borra, E. et al. (2008) Climate Sceptics on the Web, DMI case study, University of Amsterdam

Weber, I., Garimella. V.R.K. & Borra, E.K. (2012a). Mining Web Query Logs to Analyze Political Issues. In Proceedings of ACM Web Science 2012. Available at Accompanying tool online at Political Search Trends, Yahoo! labs.

Weber, I., Garimella. V.R.K. & Teka. A, (2012b) Political Hashtag Trends, Yahoo! Labs.
I Attachment Action Size Date Who Comment
anti_acta_network.pngpng anti_acta_network.png manage 571.5 K 24 Oct 2012 - 15:06 NoortjeMarres  
pro_acta_network.pngpng pro_acta_network.png manage 200.7 K 24 Oct 2012 - 15:06 NoortjeMarres  
Topic revision: r5 - 31 Oct 2012 - 13:06:16 - NoortjeMarres