Using sentiment analysis to predict opinion inversion in Tweets of political communication

Yogev Matalon, Ofir Magdaci, Adam Almozlino, Dan Yamin*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

31 Scopus citations


Social media networks have become an essential tool for sharing information in political discourse. Recent studies examining opinion diffusion have highlighted that some users may invert a message's content before disseminating it, propagating a contrasting view relative to that of the original author. Using politically-oriented discourse related to Israel with focus on the Israeli–Palestinian conflict, we explored this Opinion Inversion (O.I.) phenomenon. From a corpus of approximately 716,000 relevant Tweets, we identified 7147 Source–Quote pairs. These Source–Quote pairs accounted for 69% of the total volume of the corpus. Using a Random Forest model based on the Natural Language Processing features of the Source text and user attributes, we could predict whether a Source will undergo O.I. upon retweet with an ROC-AUC of 0.83. We found that roughly 80% of the factors that explain O.I. are associated with the original message's sentiment towards the conflict. In addition, we identified pairs comprised of Quotes related to the domain while their Sources were unrelated to the domain. These Quotes, which accounted for 14% of the Source–Quote pairs, maintained similar sentiment levels as the Source. Our case study underscores that O.I. plays an important role in political communication on social media. Nevertheless, O.I. can be predicted in advance using simple artificial intelligence tools and that prediction might be used to optimize content propagation.

Original languageEnglish
Article number7250
JournalScientific Reports
Issue number1
StatePublished - Dec 2021


FundersFunder number
Koret Foundation


    Dive into the research topics of 'Using sentiment analysis to predict opinion inversion in Tweets of political communication'. Together they form a unique fingerprint.

    Cite this