The experience of classifying the social position of the victims of political repressions in the USSR using the support vector machine

Abstract

The article describes various approaches to the classification of occupations in historical research, using the example of the database "Victims of political terror in the USSR". A brief overview of the methods by which this problem was previously solved is given: from manual assignment of certain occupations and professions of the repressed to different social groups that existed in the 1930s in the USSR, to automatic clustering. Further, a new method is proposed: to apply supervised machine learning for classification: use records already divided into groups during the author’s previous studies for training the algorithm and automatic labeling. The best of the tested methods turned out to be the support vector machine, which showed an accuracy of 95% on the test sample. The advantages and limitations of such a classification are considered, with the main limitation appears to be that some social groups are systematically defined more poorly. Nevertheless, the application of this technique made possible to mark up 350 thousand new records from the database extremely quickly. Markup based on the "training" data processed by the historian appears to be a promising direction for historical data science.