Identifying influential users in social networks is the holy grail for a whole host of enterprise activities, ranging from product marketing to product design. If it were possible to pick the top influencers from the millions of social network users, marketing would be more efficient, trends could be spotted faster, and opinions could be captured more cogently. Current approaches that try to determine influence in social networks are limited to tallying connections, posts, and followers, often without scientific rationale.

The authors of this paper propose a graph model that represents the relationship between related posts in a more complex fashion than simply counting them. The new twist in their approach is to go beyond investigating the people who start a post by stipulating a go-between, a connector that bridges clusters of posts. Distinguishing starters from connectors enables the model to capture both the explicit and implicit relationships between posts. Consequently, it identifies the most influential users from a dynamic interaction point of view, using a much more stringent method than the static endeavors that attempt to generate “vanity scores” in order to sell user data to marketing departments.

To recap the relevant definitions: Starters generate more “inlinks” than “outlinks” (that is, they receive more links than they make), which is somewhat counterintuitive to the notion of a starter. Connectors link starters, in this framework. In the proposed graph model, the edges represent actions and the nodes represent actors. Influential posts are determined by concentrating on starters and connectors.

The core sections of the paper discuss graph modeling of online postings, including graph construction, graph transformation, and influence measurements (for example, degree measure, shortest-path cost measure, and graph entropy measure). To test their assumptions, the authors conducted an experiment on Twitter that showed different results for each of the three measurement types, but in a complementary fashion. A composite view of the measurement results identified the influential starters and connectors.

While the paper presents a quantitative view of starters and connectors, a valuable extension of the work would be to look at the text components more closely to see if it is possible to identify connectors and starters via text mining techniques. The authors suggest this at the end of their paper. Another interesting study would be to compare their approach with Christakis and Fowler’s [1], who successfully predicted influenza outbreaks at Harvard using the friendship paradox [2]. This paradox states that, on average, most people have fewer friends than their friends have, so the people with the most friends should be the most influential.