Abstract

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.

中文翻译：

---

以社交网络Vkontakte为例的俄罗斯用户政治偏好建模

摘要

提出了两种机器学习模型，用于基于数据分析的微观方法自动预测Vkontakte社交网络的俄罗斯用户所持有的政治观点。结果在各种科学和应用领域中进行了测试。其中一项是监测公众舆论：基于对2200万成人用户帐户数字指纹的抽样测试，做出了两项估计，以反映用户在2018年总统大选前夕的政治偏好。使用这些估计时为了对选举进行回顾性预测，平均绝对误差分别为12％和19.4％；此外，在对前三个候选人进行排名时，第一笔估算是正确的。此外，提出了一种方法来校准模拟意见动态的数学模型的参数，即确定用户自己持有的意见的数量。该方法基于所构造算法生成的估计。