Transformation of Tweets’ Sentiment in Political Communication; The Inverted Opinion Phenomenon

Abstract

Social media networks have become an important tool for sharing information in political discourse. One of the most influential social media networks is Twitter, which boasts a simple platform and a community of over 300 million monthly active users. Previous studies examining opinion diffusion have not accounted for the observation that some users may invert the content of a message before disseminating it, propagating a contrasting view relative to the original author. Using politically-oriented discourse relating to Israel, we showed that this “Inverted Opinion” phenomenon is highly common. From a corpus of approximately 716,000 Tweets related to Israel, we identified 6,174 Source-Quote pairs. In 41% of these pairs, the sentiment polarity of the Quote text toward Israel differed from that of the Source text. Using a Random Forest model based on the Natural Language Processing features of the Source text and on user attributes, we could predict whether a Source will undergo Opinion Inversion (OI) upon retweet with an ROC-AUC of 0.83. Roughly 80% of the factors that explain OI are associated with the Source’s sentiment toward Israel. Although content with a strong positive sentiment was found to be the most viral, it is also the most likely to be inverted. Thus, it might be better to create messages expressing weaker sentiment. Our work demonstrates the importance of explicitly accounting for OI in the diffusion of political discourse.