Instrumental Methods of Media Space Analysis in Digital Humanities

The article deals with the analysis of the place and the role of instrumental mathematical methods in the methodology of the digital humanities. The authors explore the potential of those methods in terms of overcoming the fragmentation of the digital methodology due to the breadth of applicability of the method. The "sufficient" breadth of instrumental methods, which can strengthen the positive convergence of digital methodology, is ensured by their applicability to various aspects of news flows of the media space. A news stream is a collection of news generated by a) news agencies, b) preliminary materials from primary sources, c) social networks. The authors consider analytical tools as a kind of machine learning systems used to determine trends in the media space. Their applicability is focused on the quantitative assessment of text messages (nature, influence, relevance, novelty), as well as the forms of sentiment analysis of varying degrees of complexity, allowing to reflect the context of a news message, positive, negative or neutral. To do this, either the calculation of the DISAG index of the discrepancy in the evaluation of the message is used, or such machine learning models as the naive Bayesian classifier, the logistic regression, the compositions of decision trees, the fully connected neural networks, the convolutional neural networks, the recurrent neural networks. The authors pay special attention to auxiliary databases - dictionaries, lexicon and grammar, as well as libraries of subroutines designed to perform tasks related to text analysis, and aggregators of the news stream. The authors come to the conclusion that a deep analysis of the quantitative characteristics of certain news streams or interactions of the users of social networks allows solving typical tasks in the main areas of digital humanities, thereby contributing to the unification of its methodology.