The power of prediction with social media

What is it about?

Can social media predict the future? A growing number of academics are investigating, with varying success. Some have studied whether social media can predict election outcomes, or the highest grossing box office movies. Could it be used to predict movements in the stock market? There are huge gains to be made from developing a way to analyse social media and predicting outcomes to these questions. This paper gives an overview of studies done to date and discusses the various models put forward by other researchers. Types of forecasting models • Prediction market models: people give their best guesses for an outcome, then the guesses are aggregated and the aggregate guess closely matches the outcome • Survey models: people are questioned about how they intend to act • Statistical models: looking at outcomes in similar situations and analyzing the variables – for example, the length of incumbency could be used to predict the vote share of the party at the next election. Which is best for social media forecasting? Each of the above models has their advantages and disadvantages if used in the context of social media. Prediction models should work as there is a large and diverse group of potential participants. There is a risk of tampering however, which should be reported on in any results. Survey models are easy to implement, but the self-selection nature of surveys suggests the results may be biased. Statistical models seem to be the most reliable for social media data, although implementation is more complex and the model should be repeatedly tested for suitability before launch. The authors list other potential issues specific to social media, such as noise and bias. Ultimately the authors warn this body of research is in its infancy and lots more work is needed, but give some interesting accounts of where social media forecasting has been attempted and what happened.

Featured Image

Perspectives