Can social network data predict individual behaviour? And does “going viral” automatically translate to marketing success? Earlier this month, Dan Goldstein, author and principal researcher at Microsoft Research in New York City, visited Maastricht University (UM) to answer questions concerning the true value of social network data.
During his guest lecture, “Online Social Network Data: What is the Economic Impact?,” Goldstein discussed whether it is possible to predict the behaviour of an individual and, if so, with what information. According to him, a good place to start when trying to guess what someone will do is to look at their past behaviours on social media and also at what their social media contacts do. Especially if there is little information about a target’s past behaviour available, knowing what his or her social media contacts do is particularly valuable. This is supported by a phenomenon called “homophily,” which assumes that people in social contact are similar on various levels, and will therefore have similar behaviours.
Does that mean that online social network data is a powerful tool for market researchers? Not necessarily, according to Goldstein, who pointed out that as our knowledge about an individual becomes more detailed, the usefulness of social data declines. In addition, having social media data on specific individual purchases a target has made does not seem to add much to our understanding of a person’s behaviour.
Furthermore, Goldstein says that when content “goes viral” on the Internet it is not always an indicator of successful marketing. If you’re active on social media, at some point you have come across content said to be viral that has been posted by your friends. It’s easy to assume this means the content is wildly popular. However, Goldstein explained, on the Internet, what we tend to call “viral,” in fact is usually not viral from the scientific point of view. Studies presented by Mr. Goldstein proved that out of millions of diffusion trees that graphically represent the spread of information in the Internet, the vast majority of them were small, not multi-stage, and adoptions were close to the source.
For marketers, this would mean that, although the content have reached considerable number of people, the spectrum of diversity within the group is narrow which in turn means that the sample is not representative of the targeted population. Using such findings may lead to misleading outcomes in further research.
In order to support that outcome by an example from our everyday life, one may have a closer look at Twitter. Posts shared by celebrities usually get thousands of retweets. However, after reaching the first level of spread, they are most likely to stop at that point, as they don’t get any retweets or only a few from the second-level source. This is why, contrary to popular belief, such content is not viral but simply popular.
Goldstein explains more about the value of online social network data and the mechanisms behind its diffusion in his papers . Nevertheless, all those who are interested in that topic and would like to deepen their understanding should look at Mr. Goldstein’s publications: “Predicting individual behavior with social networks”