Deriving personality structure from natural language using algorithm word2vec

This thesis presents a possible new method for inquiry into the structure of personality within the lexical approach. In the first chapter of the literature review, I define the most important theoretical constructs for trait research, lexical hypothesis and five factor model. Then I summarize the current findings in the area of information technology application to the study of personality structure, concretely to its prediction, prediction of its impression and its simulation. In the empirical part I describe how to devise general personality structure through the study of correlation between vectors representing meanings of words instead of correlation of self-report questionnaires. Successfulness of this method is then evaluated using confirmatory factor analysis, cluster analysis and illustrated with the aid of several graphs. According to my results, the proposed method provides new source of information about the structure of personality descriptive words. In the discussion section I then list several possible shortcomings of this approach and propose agenda for further research. Keywords lexical hypothesis, big five, word2vec, personality structure, word embedding