Using Machine Learning and Visualization Tools to Monitor National Media

Abstract

The paper describes software automating the exploration of national media sphere. The research material is represented by texts about 207 media events reflected in the main Russian media during a year. The research was conducted using the database of MirTesen social network with help of original web application, which enables automated classification of news content. 1,02 million news articles were processed related to the media events under consideration. The processing was based on machine learning methods (supervised learning), expert analysis network analysis, and visual analytics. It is assumed that the media event is represented by the set of connected concepts of national concept sphere. Media event is a natural (however, temporary) clue to consolidate the fragments of national concept sphere, in which content and structure are of great social importance. Often different media events have shared concepts and therefore clustering is possible. Moreover, the local concept sequences, which represent individual events, are then connected into large network representing national concept sphere formed in the mass media during a certain time period. The research result is the set of media event classes with their hierarchy (conceptual media sphere priorities), as well as the model of media events interconnection.