Measuring Influence on Twitter: Top 10 Influencers on #cdnpoli

During this election campaign, a lot has been written about the influence of social media in general and the influence of twitter in particular. But who exactly are the people that actually influence and control the twitter conversations about Canada’s upcoming election? And how can we measure influence on twitter? For the purpose of our analysis, we define influence as moving people to act in a concrete and visible manner such as commenting, clicking, replying, retweeting or recommending. If you already have a specific Twitter user in mind, you can use web tools like twitalyzer.com or klout.com to assess their influence. These tools are relatively robust, all-purpose tools that give you a general idea of a tweeter’s influence within his or her own network. But one of the limitations with these tools is that they focus on the current number of followers and the followers of the people who follow a specific user. Implicit in this model is the assumption that your followers and the followers of the people who follow you are interested in most of the things that you have to say on twitter. But such an assumption is a bit simplistic for our purpose, especially when we are talking about a topic as divisive as politics.

Our task is actually a bit more complicated since we don’t really know who are the most influential Tweeters on #cdnpoli, other than possibly the five party leaders. To help answer this question we used Netlytic – a web-based text mining and social networks discovery software package that we are developing at the lab to collect and analyze tweets, emails, forums, blogs, and chats. In total, we collected a sample of over 28,000 messages (100 tweets/ hour), posted with the “#cdnpoli” hash tag between March 29 and April 9, 2011. We then imported our sample into ORA, a social network analysis software package developed by Carnegie Mellon University, to visualize this network (see below).

The nodes (dots) in this visualization represent twitter users in our dataset and lines between the nodes represent who retweeted, replied to, or mentioned whom on twitter. These types of networks (commonly referred to as communication networks) have been shown to be a better method for revealing and representing social interactions on Twitter, rather than simple ‘who follows whom’ networks [1-2]. Furthermore, by relying on communication networks, we can eliminate the presence of most fake (paid twitter followers bought from services like buytwitterfollowers.org and usocial.net) and inactive accounts since they are less likely to appear in any online discussions. And as a bonus, communication networks can also help to identify and reduce the presence of spam-type accounts because messages from spammers are less likely to be retweeted.

The retrieved network consisted of a large dense group of ~5,000 active twitter users with about 13,000 connections between them. Users found in the center of the network have the most connections. The high density of this network suggests that information posted will quickly spread to other online users of this network, especially if this information is posted by somebody in the center of this network.

Next we turned to two popular measures of centrality commonly used in the field of Social Network Analysis to find influential users:

  • In-degree centrality – the number of times a user is mentioned or retweeted. This measure allows us to find celebrity-type users or users who contribute a lot of original content to the network and whose content is frequently retweeted.
  • Out-degree centrality – the number of times a user mentions or replies to other people who also participate in a communication network. This measure allows us to identify people with a high awareness of what is going on in the network and who also tend to retweet messages by others.

Ideally, to be a top influencer, a user would score high on both of these measures. Therefore, for our ranking of twitter users on #cdnpoli we decided to rely on the combined value of the two measures mentioned above called “the total degree centrality”. To observe possible changes in ranking over time, we split our sample into four distinct time periods of 3 days each. The chart below represents the top 10 influential twitter users in four consecutive time periods.

It is not surprising that the leaders of the four national parties appear in the list of top 10 influencers on #cdnpoli. Stephen Harper was in 1st place position across all four time periods. Michael Ignatieff was in second place during the first three time periods, but dropped to 4th place in the last time frame. Jack Layton and Elizabeth May experienced higher fluctuation in their influence rankings over the observed time periods. They started in the 4th and 3rd positions at the beginning of the campaign, but moved down to the 6th and 7th during our last scan. Needless to say, the main reason why the four leaders appeared in the top 10 is not because of their active participation on twitter, but because they are often mentioned by other users and their messages (or messages about them) are retweeted frequently.

Interestingly, if we looked at the ranking of the last period under analysis (April 6-9), 3 out of 4 leaders lost their standings in the influence ranking and were pushed down by accounts owned by private individuals or organizations. Some of the more active personal accounts that appear in the top 10 list include CometsMum and ABC_now (both are critics of the conservative party with a liberal leaning), NoAPatrickRoss (a supporter of the conservative party), and jimbobbysez (a supporter of the Green party). From analyzing the tweets of a few of these individuals and organizations, one can see that these tweeters are not simply just retweeting the parties’ official positions and announcements or parroting mass media news articles. This suggests that twitter is a very open and democratic medium and that even private individuals can commandeer and steer the direction of a discussion on twitter.  This is an example of small “d” democracy in action and how a new technology can help to amplify and give a voice to the average person.

Another group of users in our Top 10 chart are reporters. For example, AntoniaZ (Antonia Zerbisias) – a Toronto Star columnist and RobertFife (Robert Fife) – the CTV News’ Ottawa bureau chief and executive producer of CTV’s Power Play and Question Period. Their tweets were often retweeted or were used to either validate or invalidate a particular point of view. Finally there were also some accounts that don’t represent any particular views such as a Canadian’s news aggregator (natnewswatch) and a “non-profit and non-partisan” organization for “sharing ideas and information that promote Canadian democracy” (DemocracyCanada).


(View the network on Slideshare)

Lastly, we visualized and examined the twitter communication network for the period between April 6-9* to find out why no supporters of the NDP made it into the Top 10 influencers list, while each of the other three national parties (Liberal, Conservative and Green) had at least one supporter (not including the party’s leader) in the Top 10. This was a bit surprising to us since Jack Layton’s account came in 6th place overall in the ranking (just above Elizabeth May’s account), one would expect that he would be more central in the network. But as you can see from the network visualization above, the NDP leader appears to be on the periphery of this network. Upon further investigation, it appears that most of the users who are connected to @jacklayton are not central nodes in this network. This in turn suggests that the NDP leader is central mostly within his clique of supporters and is not being mentioned or retweeted by other influential members in this network. This observation, coupled with the results of a recent poll by Nanos Research which showed that the NDP had lost about 6% of their support nationally since March 15, suggests that the NDP message is not getting any traction outside of their traditional base and that they might have to reconsider their campaign strategies and message.

In sum, thus far it seems that the influence ranking as measured by the communication networks of tweeters on #cdnpoli does closely reflect traditional polling numbers and trends. With more influence-ranking data collected from future election-related tweets, we believe that we can positively identify front runners and laggards in a tight election involving multiple candidates. (But we will need to have more data before we can confirm these types of observation with a high enough level of certainty.) It will be very interesting to see whether the influencer ranking of partie’s leaders in this network will change after the TV debates this week. Stay tuned for our continuing analysis of how the election is being played out Twitter and other social media platforms.

* Gilles Duceppe did not make it into the Top 10 influencer ranking, however he was added to the visualization above to show his position within the communication network relative to the other party leaders.

** One of the main reason why we only looked at #cdnpoli  and  not #elxn41 was because based on our sample, there was a 70% overlap between the two hash tags.  And also until very recently,  the most popular hashtag about Canadian politics  was #cdnpoli.  It was only recently (within last 7 days) that #elxn41 took the lead in total tweets.

*** Philip Mai contributed to the writing and analysis for this post.

Footnotes:

[1] Bernardo A. Huberman, Daniel M. Romero, & Fang Wu (2009). Social networks that matter: Twitter under the microscope First Monday, 14 (1-5), Full-text

[2] Gruzd, A., Wellman, B., and Takhteyev, Y. (2011). Imagining Twitter as an Imagined Community American Behavioral Scientist, Special issue on Imagined Communities, Full-text


5 COMMENTS

  1. Many people use the #Elxn41 hashtag instead of #Cdnpoli. #Cdnpoli is for general topics, #Elxn41 is specifically about this election. The #Cdnpoli is assumed.

    I know some people try to squeeze them both in but they’re needlessly giving up valuable characters to do so.

    It’s like in the early days when you had #Cdnpoli and #Canpoli, and some people used both for no good reason: if I’m that interested in Canadian politics I’ll search both if both are common. And if I’m interested in Election 41 I’ll search #Elxn41

    I’d be very interested to see how your numbers change if you run #Elxn41 instead

    • Good question about #Elxn41… one of the main reasons why we only looked at #cdnpoli is because based on our sample, there is a 70% overlap between the two hash tags.

Comments are closed.