Computational Social Science |
In my analysis of a semi-normal week on political Twitter I now take a more detailed look at the user network for the topic of Angela Merkel.
In this blog series, I look at a semi-normal week on political Twitter in Germany, as thorough as my day-job will allow. The topics are Angela Merkel and Ursula von der Leyen, because of they are of similar generality, but focus different things: Ursula von der Leyen recently has become a more European topic while Angela Merkel is a mirror of the day-to-day goings-on in political Germany.
I gave a broad overview in the first post There I noticed that the conservative right sphere is composed of different users that exhibit different importance. In contrast to the von der Leyen topic, Angela Merkel and subtopics were discussed by conservative right-wing influencers rather than official accounts of politicians of the German AfD (Alternative für Deutschland; right-wing, euro-sceptic party).
In the next weeks, I will take an in-depth look at the content of the tweets and how they are used in the different communities. First, we need to know about the communities, which we will take a look at today. Because the post would get very long, we do one network per post: this time, we analyze the user network for the topics around Angela Merkel. The timeframe is 8th of July 2019 to 18th of July 2019. If you like analyses like this, you can help me out by donating an access token (not the log-in).
I generated this network out of the Mentions, Quotes and Replies. Names of accounts are only displayed for verified accounts or those that are mentioned, quoted or replied to more than 400 times (meaning: accounts that have more than 400 incoming connections). The size of the account nodes indicates this quantity - how often are users interacting (unilaterally) with this account?
The colors stand for communities: users that talk more to each other than to users outside this group. This is not an objective categorization, it is broad approximation based on an algorithm that categorizes along some set of criteria. Same goes for the position of the user nodes: An algorithm computed the optimal visual representation given some criteria to optimize and the actual network structure.
To give some broad summaries for behavior, I am using these colored categories. I give names to these clusters, so as to be able to distinguish them better. The names are no objective descriptors, of course. They are merely mnemonic devices that are remembered more easily. I tried to choose these names based on the users in the cluster, not the presumed topics discussed there. Sometimes, I couldn't avoid topical names. Please stay critical when looking at these names, they are just a memory aid.
The clusters I looked at are just a sample: clusters with less than 400 users or a very low tweet output I did not give a name. This cutoff was a pragmatic decision and it is entirely possible that smaller clusters have impact.
Based on the names, you can make comparisons of the tweet output. This we will do later, first, we look at the categories themselves. I did not translate the cluster names from German, maybe I might do that later:
|afd_nah||AfD politicians, sympathizers etc.|
|regierung_und_co||government speaker and other government accounts, government politicians, media like tagesspiegel etc.|
|maassen_chebli_linksgrün||Hans-Georg Maassen (former President of Federal Office of the Protection of the Constitution), Sawsan Chebli (SPD) and left and green oriented accounts|
|satire_kritik||Satirically oriented accounts, but also general, more left-wing criticism|
|spd_und_co||SPD politicians mainly, but also other accounts like Björn Höckes (AfD)|
|springer_zdfheute||Springer media (boulevard press), but also others like zdfheute (public service)|
|qualitaetsmedien||Established and renowned news formats like FAZ, SZ, Spiegel etc.|
|breite_politik||Accounts of political parties and associated politicians|
|sonneborn_extra3_und_co||Martin Sonneborn (chief of satirical political party "The Party"), extra3 (satirical television format) and other satirical formats|
|oesterreich||Sebastian Kurz as well as other parts of the Austrian twittersphere|
|öffrecht_fernsehen||Public service television|
|umweltschutz_u_kritik||Environmental activists and organizations, but also those that are criticized by them (Peter Altmaier, Julia Klöckner)|
|yt_verschwörungstheorien_u_reaktionen||YouTube, conspiracy theorists but also activists and people engaged in refugee help|
As I said before, these categories are not objective and sometimes they don't make much sense: Björn Höcke (AfD) together with the SPD? This is because the categories are based on the engagement of the accounts with each other, not the political alignment. It is hard to put names on something abstract as engagement, so, to reiterate, these names are primarily mnemonic devices. We will see more detail about cluster composition when we look at the actual content that is passed around within and between them.
Of course, the categories are composed of many more accounts than those that are named: There are many user that are active, but only once or twice.
The account of the Springer media newspaper "Welt" you can see what I mean:
Users of different clusters interact with welt, most of them unilaterally. You can assume with some confidence that these clusters are composed of users with differing philosophies and attitudes, on average. For future posts, it will thus be interesting to see how the same news items are parsed within these clusters.
But political figures themselves are also assigned to clusters that do not make much sense on the surface. Such as the account of Hans-Georg Maaßen:
HGMaassen belongs to a cluster with many other politicians from the German left and green sphere, as well as the SPD delegate Sawsan Chebli. Maaßen is a nationalist with strong pro views on border control, so maybe he is in this cluster, because his recent tweets incited criticism. He is also cited by the cluster "afd_nah" that has many equally nationalist, conservative accounts. There is probably not much criticism here.
But we can look at hundreds of individual accounts, but an overview we will not get this way. First, let's ask: How active are these clusters?
In this table, you see the magnitude of the tweet output of these clusters within the observation timeframe (for the topic of Merkel). The number of tweets was computed for each hour. That "afd_nah" has 5980 users and produced 40.73 tweets on average means that, every hour, there were on average 41 tweets from this cluster. The median tells you how many tweets were posted by a cluster at minimum in 50% of the hours.
|Klasse||Anzahl User||Anzahl Tweets||Median||Mean|
You can see very clearly that the cluster "afd_nah" is much more active than all the other clusters. It is the only cluster has a median of 1, the baseline activity is higher here than for any other cluster. The mean is also more than two times than that of the cluster with the second highest output. The relationship of number of users to number of tweets is also different: While only 1.5 times more users than "regierung_und_co", the number of tweets is three times as much.
This was to be expected: The AfD and its sympathizers understand themselves to be systematically ignored by the mainstream media. Going by this logic, it makes sense to be louder than anyone else.
The relationships between the number of users and number of tweets is more regular for the other clusters. We can assume that these clusters outputs fluctuate more regularly, influenced by the daily flow of news and information. Possibly, the fluctuation of output for the cluster "afd_nah" is less influenced by actual events in the news. But this has to be shown in future posts.
In general, clusters that have accounts of high reach and relevance produce more output. This is also not surprising, because news or government accounts regularly produce content that is controversially discussed or at least highly retweetable.
Can we look at how much the clusters have produced over time? Yes!
These plots show the activity of the users within these clusters at each timepoint (each hour in the timeframe). You see very clearly that the baseline of user activity activity is much higher for the cluster "afd_nah".
As mentioned, "Angela Merkel" is a pretty general topic, she comes up in many political day-to-day discussions. Because of this, activity flows with the flow of news. In the week of 8th of July to 18th of July, this was mostly influenced by the election of Ursula von der Leyen to President of the EU Commission on the 16th, as well as the announcement of that election on the 11th. On the 11th, Angela Merkel also made a statement about her public tremblings.
The higher baseline for "afd_nah" is possibly explainable by a tendency to discuss general, hard-to-verify perceived grievances. A frequent "luring" with news items might not be necessary. In any case, "afd_nah" is not as influenced by the mainstream news cycle.
This influence is easily seen for "sonneborn_extra3_und_co". Around the 16th, this more satirically oriented and europolitically oriented cluster has gotten much more active.
This is the same graphic without the less active clusters, for your convenience:
It is also interesting to look at the same graphic for the mean of user activity. Here, you can see that users are not as different on average. Even for "afd_nah", most users produce next to nothing (median is always 0). The drag the average down to elevated yet normal levels. The divergence between highly active and minimally active users might be greater here. Interestingly, this is not the case for "sonneborn_extra3_und_co", showing that users here have possibly similar activity patterns (which you can also see in the table above: there is less of a difference between number of users and posts)
This does not tell us that much about behavior: Who is talking to whom?
Below, there is another network. This time, the interactions between and within clusters were summarized on the cluster-level. Arrows between the clusters indicate interactions between users of these clusters. The thickness of the arrows shows the relative intensity of interaction: it shows how much of their general output is aimed at the different clusters (Mentions, RTs, etc.). So, the thicker the arrow, the more clusters interact. The size of the arrowheads indicates how much of the output a cluster cites that specific cluster. The cluster "yt_verschwörungstheoretiker_u_reaktionen" cites "afd_nah" way more than the other way around.
The size of cluster nodes indicates the percentage of output a cluster allocates on itself: how much is a cluster talking to itself relative to talking to others?
My initial expectation was that "afd_nah" would be the biggest bubble. That's not true, at least not for this topic and timefram, the clusters "satire_kritik and "sonneborn_extra_3_und_co" cite themselves even more. "österreich" (Austrian twittersphere) is also relatively self-focused, but this is not a surprise at all (it being a different nation)
Converning the relations, there are not many trends visible, but some are prominent: The connections of "afd_nah" to the satirical-critical, left or green clusters are very slim, meaning the active accounts and audience is different and does talk to each other less. "afd_nah" does interact with government or media focused accounts, but it has to be said that accounts from the AfD are also present in "breite_politik".
You can also see very well, that "yt_verschwörungstheoretiker_u_reaktionen" are mostly unilaterally outgoing (citing different clusters, but not other way around). This is more or less explainable by the endemic nature of many of the majority of these accounts, that I found after a brief examination: people interested in things from right-wing politics, political conspiracies, to chemtrails and lizard people. There are also accounts engaging with refugee help and YouTube itself. It could be, that conservative and conspiracy-interested accounts simply are the majority, but YouTube as a commonality "binds" them together.
Also prominent are the unilateral, outgoing relations of "österreich" to many other clusters. Unilateral means that relatively, "österreich" cites other clusters much more than the other way around. Some stronger connections are present to political and media clusters, but also "afd_nah".
This could be an artefact going back to the categorization itself: Many AfD politicians and sympathizers are also present in other clusters (clusters being based on engagement, not political alignment).
Yet, polarization is clearly discernible and I ask myself if this will be the same for the discussion on Ursula von der Leyen and associated topics. As established in the first post, there are different audience compositions and Merkel is more of a stable topic than the more "event centric" topic Ursula von der Leyen.
Next time, I will look at the network for Ursula von der Leyen. The following weeks will then look at content and how different clusters deal with different topics. For questions, suggestions or critique you can reach me under @Boomcrashkapow or firstname.lastname@example.org schreiben.