In recent years, researchers and marketers started showing more interest on whether the diffusion of ideas can be maximized by “seeding” information or a new product through special individuals called “influencers”. Influencers are people who have some combination of attributes –such as credibility, expertise, or enthusiasm, or network attributes such as connectivity or centrality –that enables them to influence a large group of people, via a cascade of influence.
This study uses Twitter to understand the role of influencers in the diffusion process. Given that Twitter is devoted to spreading information with users following one another, this analysis offers a look into the network of “who listens to whom”, or the “ follower graph” in other words. The paper studies the attributes and relative influence of 1.6M Twitter users by tracking 74 million diffusion events that took place on Twitter over a two-month interval in 2009.
The results, unsurprisingly, confirm the general notion that the largest cascades tend to be caused by users who have a high number of followers and have been influential in the past. However, this observation is paralleled by a more surprising finding, that large cascades are relatively unreliable. In a way, word-of-mouth diffusion can only be generated reliably by targeting large number of potential influencers, creating average effects.
In order to calculate an “influence score”, the study tracks the URL posts from their origin at a particular seed node through a series of reposts. This helps us to understand how much the links that people post are diffused in their network until the cascade is terminated. If person B is following person A, and person A posted the URL before B and was the only of B’s friends to post the URL, we say person A influenced person B to post the URL.
Of course, there is some difficulty in assigning influence to people. Some of the challenges arise when we acknowledge that there can be situations in which credit is shared among different people. If influence is defined over a chain of people sharing the post before it spreads more, we can either assign the influence to the person who posted it first, or the second person who propagated it, or even the last person who posted it most recently. Alternatively, we can split credit equally among all prior-posting friends.
These assignments make different assumptions about the influence process: the notion of “first influence” emphasizes primacy and assumes that the individuals are influenced when they first see a new piece of information. Alternatively, the “last influence” notion instead attributes influence to the most recent exposure. Lastly, the “split influence” approach explains that one person’s inclination to act on a piece of information increases as he/she is exposed to it multiple times, creating an accumulation of influence. One footnote to this explanation is that even though it is easy to explain why people share these posts through theories of influence, we should also pay attention to the homophily factor. It is indeed likely that people who are friends or follow one another are similar in many ways and have access to similar sources of information.
In light of this information, this study carries out an economic analysis of the cost-effectiveness of marketing strategies (compensating prominent individuals vs. identifying average/ordinary people with potential to be influential). The result indicates that the most cost-effective performance can be realized using “ordinary influencers”— individuals who exert average or even less-than-average influence. These findings suggest that instead of trying to identify extraordinary individuals, marketers should adopt portfolio style strategies, which would target many potential influencers at once and rely on average performance.
This result contradicts the general emphasis placed on using prominent individuals as the optimal market strategy in disseminating information. Therefore, the paper concludes that marketers interested in using word-of-mouth influence could benefit from adopting more precise metrics of influence, collecting more data about potential and using ordinary influencers.
(Link to the article)