This is the first of a series of posts in which I will present analysis on data collected on the networked-public sphere surrounding the ongoing battle for Fallujah. I’ve been messing about with some data mining, analysis, and visualisation techniques and thought a comparison of how the English vs Arabic language twitter spheres have responded to the Fallujah operation might be an interesting case to test them out on. In this first post I will keep the discussion mainly at the level of the entire network by analysing the basic network structure and the information and narrative framing of the discourse circulating within it. In future posts I will look more closely at the relationship between network structure and the production of discourse. To do this analysis I used a combination of NodeXL and Gephi.
Fallujah and the Networked Public Sphere: the English language sphere
The first data set I am going to explore consists of tweets collected between the 7th and 9th of June using the search parameter ‘#Fallujah OR #Falluja’. The data consists not of tweets per se but interactions between users i.e. when a user mentions or replies to another user/users. Such interactions generate social ties (edges) between nodes (vertices) which represent the twitter user. Because of Twitter’s data retrieval limits, it is not possible to simply bulk request an entire network around a particular search term. Consequently, this data is only a sample of the total, generated by repeated periodic requests to the Twitter API over a three day period. The final data set consists of 3320 edges and 2572 vertices (i.e. relationships and accounts.)
Step 1: Entire network visualisation
Figs. 1.0 and 1.1 show this network visualised using the ForceAtlas2 (FA2) algorithm through Gephi. For more details of FA2 click here. Figs. 1.0 and 1.1 have been coloured according to clusters and node size reflects eigenvector value i.e. the importance of the node within the overall network. Fig 1.1 strips back smaller nodes and clusters and labels the key nodes with the twitter username of the account.
The network map exhibits the characteristics of the community clusters network structure. This indicates that the discussion and flow of information around the Fallujah operation involves multiple small groups connected to key hubs which are functioning as sources of information and influence. This can often be the case with breaking news stories as news agencies become hubs for disseminating information. In this case, the key hubs are not principally news agencies but ‘activists’ who are functioning both as sources of scarce information on the ongiong operation and to shape the narrative of events in quite radical ways as will be discussed below.
Each cluster within the network can of course be characterised by a distinct structure, as is the case here. We can see immediately that the account @moonnor27 is an influential hub in this network. However, the cluster associated with this node is also rather isolated from other clusters i.e. there are a high number of interactions within this cluster but comparatively little interaction with outside nodes or with other key hubs. This cluster is also narrowly focused on the single central node showing a high number of in-degrees to that account but very few edges between other nodes in the cluster. This might tell us something interesting about @moonnor27’s audience. By contrast the cluster in which the account @nidalgazaui is the most influential node shows both more internal interactions between nodes and greater external relationships with other clusters. In part II I will be delving into a more granular analysis of these key nodes and clusters and their function within the network. Here I want to present some characteristics of the discourse emerging from information shared within the entire network.
Step 2: What information is passing through the entire network, how is the discourse being framed?
The emergence and characteristics of a discourse on the Fallujah operation in the English language twitter sphere can be analysed by extracting data from the network which is amenable to discourse analysis. This data can be gathered in different ways. Firstly, by word frequency and word association measures. Secondly, by using sentiment measures. And thirdly, by calculating which urls have been most frequently shared in the network. It is important to note that this information is not answering the question: ‘Which articles were most frequently tweeted alongside the #Fallujah hashtag?’ but the more interesting question of: ‘Which articles were most commonly shared or directed between users alongside the #Fallujah hashtag?’ In other words, the flow of information between nodes is what is being measured.
What emerges from this analysis bears out the impression that in the English language public sphere the Fallujah operation has been interpreted in fairly narrow Sunni-Shi’i sectarian terms. A more detailed picture of this narrative can be ascertained by looking at the word pairings and hashtag frequency data presented in tables 1 and 2 below. Here we can see that the battle for Fallujah is being discussed most often in terms of Shi’a militias, Iraqi forces, and Sunni civilians. Interestingly, after the term ‘shia’, the term ‘militia’ is most commonly associated with ‘warcrimes’. Also worth noting is that hashtags for ‘iran’ and ‘soleimani’ feature prominently within the network. The overall picture, then, indicates that the discourse on Fallujah is overlaid with a sectarian framework which frames participants according to a Sunni-Shi’i binary, is particularly concerned with the actions of Shi’i militias which are construed in an extremely negative light, and, amongst non-Iraqi participants, is primarily preoccupied with the role of Iran.
|Top Word Pairs in Tweet in Entire Network||Entire Network Count|
|Top Hashtags in Tweet in Entire Network||Entire Network Count|
These conclusions are further borne out by an analysis of the urls being shared within the network, shown in table 3 below. The most shared article on this network is a CNN Arabic story titled: ‘Al-Abadi orders the arrest of those accused of “excesses” in the battle for Fallujah, the Association of Muslim Scholars condemns the “Iranian project”‘. This article reports on accusations of abuses by Shi’i militias in Saqlawiyah and political attacks on the Iranian role in the Fallujah operation by Iraqi-Sunni bodies. Also prominent in table 3 are two widely shared links to a youtube video reportedly showing hashd militias saying that they will take revenge upon the civilians from Fallujah. A third video linked to is a Russia Today (RT) debate featuring Lina Khatib (Head of MENA programme of Chatham House.) The RT host frames the debate with the following introduction: ‘Branded as liberation, it has already sent thousands fleeing, while encouraging others to join ISIS ranks. Is Fallujah’s so called liberation likely to cause more death and destruction than leaving it under ISIS control?’ Other urls shared are less value-laden news articles containing updates on operational developments. At the bottom of the table is a Fars News (quasi-Iranian government agency) article. This article obviously presents a quite distinct narrative from the dominant themes which are foremost in the network. It describes the participants in the battle as the Iraqi army and ‘local voluntary troops’ vs ‘terrorists’ who are said to be using civilians as human shields and forcing them to take up arms and fight on their behalf.
Taken together, these measures suggest that the discourse on Fallujah circulating in the English language public sphere has tended to be framed in a very particular way. This framing has a strong sectarian dimension which interprets the battle as pitting Iranian-backed Shi’i militias against Sunni civilians, and to a lesser extent ISIS. While this has been the dominant narrative of the network taken as a whole, the existence of a counter-narrative/s is also suggested both by the existence of the Fars News piece amongst the most shared articles in the network, and, perhaps more importantly, by the fragmented nature of the cluster structure. This counter-narrative resides in the smaller but still significant hubs that have little or no interaction with those hubs from which the dominant discourse is emanating. The existence of these hubs is evident from the spatialised visualisation, but not from the entire network statistics which limit our view to the dominant discourse. Exploring this counter-narrative/s requires looking below the entire network level to address the individual clusters directly, examining the attributes which might help explain the proximity of nodes and the structures which shape the flow of information and influence within the network. These will be explored in part II…