The paper presents initial experimental results of ongoing research into the use of
Artificial Intelligence (AI) for discourse analysis of everyday internet discussions on salient public
policy issues. The ultimate goal of the research is to develop approaches and AI instruments
helping discussants reduce the excessively polarized opinions in a deliberative and dialogic
manner. This first experiment involves training of the Recurrent Neural Network (RNN) to assess
its potential to predict attitudes of discourse participants towards the Russian government's
policy to increase the retirement age, as a case study. It was done by conceptualizing a dedicated
discourse model adapted to the machine learning needs which defined the scope of a training
data set and its annotation scheme. The trained RNN demonstrated the prediction accuracy of
The paper presents initial experimental results of ongoing research into the use of Artificial Intelligence (AI) for discourse analysis of everyday internet discussions on salient public policy issues. The ultimate goal of the research is to develop approaches and AI instruments helping discussants reduce the excessively polarized opinions in a deliberative and dialogic manner. This first experiment involves training of the Recurrent Neural Network (RNN) to assess its potential to predict attitudes of discourse participants towards the Russian government's policy to increase the retirement age, as a case study. It was done by conceptualizing a dedicated discourse model adapted to the machine learning needs which defined the scope of a training data set and its annotation scheme. The trained RNN demonstrated the prediction accuracy of nearly 90% with the training dataset of some 7,000 posts. The accuracy visibly drops as the dataset size gets smaller.size gets smaller.