Carl Miller discusses development of effective social media research for policy making during a seminar on quantitative methods in social media research held at the OII on 26 September 2012.
A team of CASM staff and experts used Facebook, Twitter and YouTube to develop (a) a predictive analytic to predict the outcome of each week's vote on X-Factor based on social media users' conversations online, and (b) a real-time visualization of the audience's reaction to each contestant as they sang. The predictive analytic modelled two underlying variables: voter sentiment and voter sediment. This is based on the psephological insight that people can vote either due to the 'sediment' of a longer-term and established loyalty for a contestant, or on the short-term 'appraisal' of their immediate performance. The sediment score combined a cumulative volume 'positive' comments on Twitter with Facebook likes. The sentiment score combined twitter sentiment (positive over positive + negative) and YouTube likes-per-view. Recursive best-fit analysis was conducted to get the best weightings for these variables. The predictive analytic made 11 predictions. Nine were correct. The engine producing the real-time visualization collected X-factor relevant tweets and sorted them by contestant. They were then classified using a natural language processing engine into positive, negative and neutral categories. Positive was divided by negative and averaged within a two-second time window to produce a candidate score and then mapped onto constantly updating graph. The Centre for the Analysis of Social Media at Demos is a research body dedicated to inform policymaking through social media research. Computer and social scientists at CASM work together to find new methods to do this that are reliable, powerful and ethical.
