Improved Twitter Virality Prediction using Text and RNN-LSTM

Abstract

The matter of influence and virality in social media has been studied since the popularity explosion of these platforms. A gargantuan amount of news and political messaging transits through Twitter every second, making it a formidable force for the propagation of information. In order to stay competitive, traditional media needs to participate in these platforms and attain influence. We propose a method to predict the influence of news tweets. To this end we use several thousand tweets to train a RNN-LSTM to classify news tweets as influential or not influential using a corpus of 5000 automatically labeled tweets according to their influence. Our method reaches an F1 of 0.83, while training and classifying in under 300 seconds.