How network analysis helps journalists identify social media influencers


Influencer marketing through social media quickly rose in prominence as more and more people joined social networking sites like Facebook and Twitter.  Today, campaigns and brands are regularly built around social media influencers, and the strategy itself is part and parcel of nearly every effort to win favor with customers, manage corporate image, and mitigate crises.

But questions remain: what good are influencers to journalists? And how can they be accurately identified?

While a considerable amount of research has examined best practices for identifying influential actors in social media, the field still remains opaque. Moving beyond obvious influencers, with millions of followers or thousands of tweets and retweets, to meaningfully reach current and prospective customers that are influential on certain topics remains challenging.  The question of how to best identify social media influencers thus persists. 

So how can journalists effectively address the ambiguity in understanding influence, particularly in a data driven environment?  Getting to know the basics of network analysis – in other words, identifying which important users connect to other users – will provide the ability to decipher exactly what influence is and how it is measured, beyond simple metrics of post activity and friends or followers. 

Image: Network graph of 1,100,011 Twitter users and 2,760,607 tweets about #Ferguson (from 11/23/2014 - 11/29/2014) summarized into 1,500 nodes and 15,173 directed edges. Node color indicates communities and node size indicates influence.

Based on numerous studies my colleagues and I have published in leading peer-reviewed journals, we outlined a methodology that moves beyond retweet or following networks. Instead, our methodology models user interactions in the form of the @mention function. You can explore one of our dynamic online graphs of influencers here. In this graph, it is crucial to note that this measure of influence is specifically based on how users actually engage with other users, and thus is not a passive measure of influence based on overall popularity but on behaviors specific to precisely identified topics.

To some extent, the use of @mention can represent not only an engagement but also a conversation. In analyzing millions of posts and unique users collected using the Boston University Twitter Content and Analysis Toolkit (BU-TCAT), we can visualize and algorithmically sort Twitter users by influence and communities.

Identifying these influencers gives journalists better insight into the users that are thought leaders on a given a topic, and expand the rolodex of potential sources and background for specific stories.

In our studies, for example on #Ferguson, we applied an algorithm called betweenness centrality that determines just how often each user acted as the shortest path to other users by mentioning them. Put briefly, this algorithm identifies which users are gatekeepers of information to other, influential and diverse users across communities of users. And, these users are the ones that can share and spread messages through a network in a highly efficient manner – and if properly engaged with by journalists, they can not only help to find new angles on stories but also help to popularize stories that are published.   

But of course, it isn’t just influence, it is also attitude. And here, the notion of communities is an important one, because social media outreach often requires connecting with users that share either favorable (or unfavorable) opinions on topics or issues. In another study, we applied a community detection algorithm to locate users that were generally supportive of certain issues as well as those that were opposed, and those that were more or less ambivalent.

Understanding influence in this sense requires attention to algorithmic sorting at both of these levels, and makes the search for relevant social media influencers far more efficient. One example of this would be political campaigning, where journalists would be especially interested in influential and undecided voters that a combination of algorithms can jointly gauge, rather than just influential users that may be from staunch liberal or conservative groups.

Building on these efforts, in a recent partnership that I have formed at Boston University with Vishal Mishra, the CEO & Co-Founder of Right Relevance, we are expanding this methodology further and at greater scale, with curated information and intelligence from more than 50 thousand unique topics from the Right Relevance archives.

Much like the process of applying sharing and community algorithms, Vishal and his team have developed a 2-level proprietary People Rank (PR) algorithm that acts as a customized sorting system for influence in social graphs. Their custom PR algorithm is based on the now-famous Google Page Rank algorithm but is tailored for social networks, as opposed to webpages and the links between them.

With the wide range of approaches to identifying social media influencers present in the field, the process can seem daunting. But, with a bit of background work dedicated to understanding how influence is measured by actions and connections, various communities around specific topics can be further subdivided into what can be considered tribes and flocks.

It is important to remember that influence is going to vary over time, even within topics, and so it is necessary to always monitor the dynamics within the networks. There are a range of analytic tools for these purposes and we unpack many leading platforms and services, including Right Relevance and Crimson Hexagon, at our upcoming Making Social Media Matter workshop at Boston University.

Enroll and earn a certificate in data science at our next workshop this October here. Additional details on Right Relevance and its People Rank algorithm are available here.