On Predicting the Success of Political Tweets Using Psycho-Linguistic Categories

Politics is increasingly influenced by Twitter communications and more and more politicians are using this platform to keep in touch with their electorate or to make proselytes. However, an effective communication is not easy to obtain and many studies are trying to identify strategies able to produce successful tweets. Day and time of publication, presence or absence of hashtags and images are some metrics used to define a communication strategy, but in this paper we propose a novel approach. Our idea is to describe tweets through novel psycho-linguistic categories (e.g. immigrants, security, kindness, etc.) and to use these categories as features to predict the success of political tweets. Through the observation of tweets posted over the last year by an Italian politician, very active and popular in the social scenario, we analyze two different machine learning approaches (i.e., Decision Tree and K-Neighbors) to predict the success of a tweet. The results obtained show that it is possible to achieve a prediction accuracy of 76%, thus showing the importance of the proposed psycho-linguistic categories to characterize tweets in the political context.