Tuesday, 5 June 2012
Social Media: human or algorithmic analysis?
Less than a year ago, I came across research by the Rensselaer Polytechnic Institute showing that once only 10% of a community hold a certain belief, that belief will be adopted by the majority. More recent research by the Network Science Center at West Point even shows that within social networks, it is possible to find a 'seed set' of less than 1% which will cause the rest of the network to adopt a specific trend.
Both sets of research use complex mathematical algorithms, which, while insightful, unfortunately miss that unpredictable human element. Given large enough populations and large enough data sets, human behaviour can be modelled mathematically fairly accurately. However, if it was that simple, planned economies would work. Users on Slashdot commenting on the most recent findings, correctly pointed out that an algorithm that finds influential nodes in a network based on measurable data such as amount of connections, may well be tracking bots or other types of bogus accounts that are ignored by the community. Not to mention that certain views are more palatable or better articulated than others. Only a minority trend that resonates well with the majority will actually be adopted according to exponential models. So far, computational models lack insight into the complexity of real-world human behaviour that is so essential to PR, marketing, and crisis management.
In crisis management, finding that 1% seed set before the message goes viral is key. Traditionally, social media monitoring only takes a closer look at the trends that get a high volume of online mentions. However in a crisis, it would be better to stop high volumes from developing at all. By influencing the right seed set, PR disasters could be averted. But it will take more than an algorithm to find the seed set you need to influence. Only a human analyst will be able to determine whether any one network node is the right one by considering how the community actually responds to the members of that seed set. For example, people may share content they disagree with in order to expose it for criticism; or they might follow an account for free goodies, but not engage with it further. Once the target group is determined, it will still take skilled marketing and PR people to package the message in such a way that it becomes both acceptable and worth advocating to the target audience. Exponential growth does not occur automatically like in the mathematical models. Persuasion is still a human skill.
Another problem with mathematical modelling is that it works best with very large data sets. Not all brands or organisations active on social media have the luxury of millions of followers or users. As a community gets smaller, statistical probabilities will be increasingly distorted by individual idiosyncrasies. For brands or organisations that are still growing a small community into a larger or more engaged one, human analysis of trends and influencers will still return the best results.
That said, the mathematical models developed at RPI and West Point are an incredibly valuable tool. Besides highlighting the importance of the smaller voice, which should inform a new and better approach to social media metrics, these types of algorithms can also help to identify areas of interest for human analysts to focus on. The ideal social media analytics offering should be a concerted approach using automated keyword searches, computer modelling of network connections, and human interpretation.