Modelling Russian users' political preferences

In this paper, we present two machine learning models that can predict Russian VKontakte users' political preferences. They imply operationing at the users-level. We consider thoroughly its different applications; one of them is public opinion monitoring. To demonstrate it, we test them on the sample of 22 mil of Russian users of age. Finally, we retrieve two estimations of public opinion. In case we value the outcome of the 2018 Presidential election by these estimations, we get MAE of 12 and 19.4 percent correspondingly. Moreover, one of the algorithms finds correctly the first three places. Another prominent utility relates to the calibration of opinion dynamics models where we can use scores generated by the machine learning algorithms to estimate users' opinions numerically.