METHODOLOGY FOR DETERMINING OPTIMAL CATEGORIES FOR CLASSIFYING A NEWS ARRAY

Goryachkin Boris Sergeevich, Korenkova Tatyana Vyacheslavovna, Chernykh Yulia Sergeevna

One of the most important sources of analysis of social events and processes is news, since they reflect almost all their aspects and afford the opportunity to build a complete picture of social reality. To do this, it is necessary to carry out a preliminary classification of news by social topics, and the original news headings in various news resources are not well suited for this task. Therefore, in this paper it was developed and tested in practice a methodology for determining the optimal categories for classifying news texts, in particular, for the social sphere. The methodology includes the definition of new preliminary news categories by the Word2Vec algorithm, multiple thematic modeling using Zero-Shot classification and semi-automatic modification of categories until the desired thresholds of the derived metric are reached. As a result, an optimal list of categories reflecting social reality was obtained, and its advantage over the initial categories was proved.

Keywords:News topic modeling, news classification, social modeling, Word2Vec, Zero-Shot classification, NLI