Predicting controversy

Predicting controversial news using Facebook reactions

Different events and their reception in different readers communities may give rise to controversy. This work proposes an entropy-based model based on the distant supervision approach to predict news controversy using Facebook reactions as a proxy.

Cross-validation and cross-source experiments have proven the validity of the approach. Furthermore, we also present an automatically generated corpus for controversy prediction in Italian.

Dataset & Code

We release (here) an accompanying jupyter notebook containing all the code needed to reproduce the results.


Send us an email to get the corpus.


We presented this work at Clic-it 2017; the slides are here. The original paper can be downloaded from here.


For bugs, feedback and suggestions please contact