Identification of Affective States Based on Automatic Analysis of Texts of Comments in Social Networks

The paper considers the problem of classifying 3553 English-language comments from thesocial network Reddit based on various approaches to the vectorization of comment texts,including bag of words, TF-IDF, bigrams analysis based on pointwise mutual information (PMI)and sentiments, and the deep model BERT of the language representation. The use of a hybridapproach based on text vectorization using BERT and bigrams analysis have made it possible toimprove the quality of comments classification up to 91%. Based on a cluster analysis of 1857English-language comments describing anxiety, clusters were identified using BERT+k-means.The study proposes a hybrid approach based on the use of the LDA topic modeling method, theVADER sentiments analysis method, pointwise mutual information, and parts of speech analysisand permitting one to select bigrams and trigrams to describe clusters of comments. To visualizethe extracted patterns in the form of trigrams, a knowledge graph was constructed that describesthe subject area, and a comparison of the words of the selected target trigrams with the words ofa custom dictionary describing various affective disorders has made it possible to determine thetypes of psychosocial stressors associated with affective disorders.