Abstract

Relevance. The designing of tools to identify a psychological distress in network is one of the most significant challenges of the era of information technology. There are evidences of certain peculiarities of the speech and textual activity of frustrated person. However, for texts in Russian, any tool for monitoring of the intensity of frustration in online content does not currently exist. 

Objective. The purpose of our work is the listing of text features to carry out automatic analysis of the network content for detecting texts of frustrated users.

Methods. The material of the study is a set of posts and comments of 2-10 sentences collected in social networks LiveJournal, Pikabu and Facebook were written by 100 Russian-speaking users aged 27-64 years. The texts were divided as written by unexcited persons (500 texts) and by frustrated persons (500 texts). For automatic text analysis, the "RSA Machine" created in Federal Research Center "Computer Science and Control" of Russian Academy of Sciences was used, which allows to determine 197 text features, to compare texts, and to identify the most important dividing features. Mathematically, the texts were classified using the machine learning.

Results. The Random Forest method with a preliminary binarization procedure revealed the most significant features of text written by frustrated person: the sentiment; the frequency of punctuation, negative word forms and first-person pronouns; the number of semantic roles causative, liquidative and destructive; number of particles, invectives and words from the vocabulary of resistance.

Conclusions. Using the identified features the network texts written by frustrated person can be confidently determined; it is applicable for monitoring in order to ensure information and psychological security.