Two models of machine learning are proposed for the automatic prediction of political views held by Russian users of the Vkontakte social network based on a microapproach to data analysis. The results are tested on various scientific and applied fields. One of them is monitoring of public opinion: based on testing a sample of 22 million digital fingerprints of adult user accounts, two estimates were made reflecting the political preferences of the users on the eve of the presidential election in 2018. When these estimates were used to develop a retrospective forecast of the elections, the average absolute errors were 12% and 19.4%, respectively; moreover, the first estimate was correct in ranking the first three candidates. In addition, an approach is presented to calibrate the parameters of mathematical models simulating the dynamics in opinions, namely, the quantities that determine the opinions held by users themselves. This approach is based on the estimates generated by the constructed algorithms.