Forecasting with Twitter data: an application to USA TV series audience

In this paper the interactions between television audience and social networks have been analysed, especially considering Twitter data. In this experiment about 2 and a half million tweets were collected, for 14 USA TV series in a nine weeks period through the use of an ad-hoc crawler created for this purpose. Subsequently, tweets were classified according to their sentiment (positive, negative, neutral) using an original method based on the use of decision trees. A linear regression model was then used to analyse the data