[important]This post is the third in a series of posts highlighting some of the research articles featured in a new Special Issue on Networked Influence in Social Media recently published by the American Behavioral Scientist (ABS) (Issue Editors: Drs. Anatoliy Gruzd, Ryerson University, and Barry Wellman, University of Toronto).[/important]
Trace data (the bits of ourselves we leave behind when navigating the Internet) is exciting.
It is exciting for academics who are permitted a unique view into the communicative processes of individuals. It is exciting for marketers who can develop better strategies based on new understandings of how people interact with their products. It is exciting for the average citizen who is now able to reflect on their social life with precision.
When we use Twitter or other online social networking tools we leave traces of our actions, kind of like footprints in the sand. We leave a little note for whoever wants to collect it (before it washes away, that is). It says “I was here.” When people like Devin and I collect those notes from an application like Twitter, we can use them to better understand who the main players in a given online community are and who is most influential within that group of accounts.
For our study we looked at #CPC and #NDP, two of the largest political Twitter communities in Canada. We wanted to identify the most influential political players and so we turned to the literature and realized, there are a lot of options! Researchers and marketing professionals alike seem to have a plethora of tools for making use of trace data but little agreement on which is best, or even what different kinds of traces mean. Things like indegree, eigenvector centrality, retweets/mentions, and quality of content are all basic metrics various others have used independently to identify influentials. We wanted to see which one works best in the context of political discussion.
To figure out the “best” one implies an end goal – best for what?
Well, traditional measures of network centrality like indegree (in this case how many followers within the community an account had) and eigenvector centrality (based on how many followers who also have a lot of followers an account has) turns out to be pretty good at identifying the traditional political elite.
Measures that consider how much interaction an account has (retweets and mentions) and quality of the message (if messages were explicitly political or not) tended to identify a different group of top influentials consisting of bloggers and political commentators.
What none of the standard measures were particularly good at was sorting out who “opinion leaders” are. Opinion leaders are average citizens who happen to care a lot about a given topic, in this case Canadian politics. They are thought to be particularly good at influencing their friends, family and everyday associates because of their shared social bond.
Check out our article in the new American Behavioural Scientist issue on identifying influentials in social media to find out how Devin and I propose to find those opinion leaders and to read more about the meaning behind trace data.