Determination of consistent patterns of suspiciousness in news texts

Abstract

This article presents the results of the analysis of the corpus of texts of Kazakhstani
media in order to identify the patterns of suspicious informative features of unreliable news. The data
obtained are based on expert analysis of 5,2 thousand news texts. The process of forming and marking
the text corpus according to the selected informative features of reliability, tonality, bias, politicization
and the presence of manipulative techniques is presented in stages. The paper also considers the issues of
public response of news and their causes identifying factors affecting the popularity of news texts using
statistical modeling. To improve the reliability of the results obtained in the course of testing on recognition
of informative features in the news triangulation method was considered and applied. According to the
results of this work, a number of consistent patterns were established for the machine learning procedure
to determine the influence of a set of informative features on the popularity of news.