Application of random forest in choosing a method for the age of a social network user recovery

This article is devoted to the problem of a social media user’s age inference. We considered methods based on the analysis of profile information about education, subscriptions and information about the education of friends. All of these methods can be used individually for samples of users with certain characteristics. To increase the proportion of users whose age can be inferred, we combined several methods in three ways: ranking, averaging the scores, and using a random forest to build a classification algorithm that selects the best method in each case. By combining the methods, we built a model with an age inference accuracy of 0.80 on a test sample. The theoretical significance of the work lies in proposition of a method for combining age inference algorithms, which increases the applicability and accuracy of individual algorithms. The results obtained can benefit the analysis of user profiles in social media in many areas.