WORD2VEC- AND SEMANTIC GRAPH MODELING OF EMOTICONS AND EMOJI USE IN INTERNET COMMUNICATION TEXTS

Abstract

The article deals with emoticons and emoji use in textual computer-mediated communication. The research material is based on comments left by users of the social networking service VK. Using machine learning methods (Word2Vec), we build a model of contextual proximity of emoji use in the comments. This model reveals a closer contextual relationship of negative-tone emoji of different types with each other. The influence of personality’s psychological parameters on the use of icons of different tonality in similar communicative situations is considered as a hypothesis that explains the greater variability in the choice of negative-tone emoji when writing similar contexts. This hypothesis was tested by using a tagged corpus of texts left by users that had passed the psychological test to determine the BFI degree. We collected automatically and analysed the comments in the information system “Semograph”. It was found that variability in the use of negative-tone types of emoticons and emoji (namely the icons with the semantics of sadness, fear, apathy and bewilderment) in the same context is more characteristic of different psychological types of personality. Conversely, positive-toned icons (namely the icons with the semantics of joy, love, thankfulness and approval) are less variable in the same context.