Computational Personality Prediction Based on Digital Footprint of A Social Media User

Abstract The digitisation process of objects and operations of the real world is quite active, i.e. creating their digital entities or models. People regularly leave enough of their data in social networks services and on various sites, thus forming their unique digital footprint. Based on the obtained digital footprint, it is possible to create a complete digital entity of a person in the hyperspace of social media. However, a person is a complex system; therefore, the model of a digital entity must be multi-scale. In this paper, the relationship of such a weakly formalizable side of the user, such as his psychometric indicators, and his digital footprint in the social network sites have been studied. First, it was conducted a series of experiments on the prediction of psychometric, based on data from social networks (Facebook and Vkontakte), during which two prediction approaches were compared: multi-response forecasting, when all psychometrics are predicted simultaneously, and univariate models for each personality trait. Then, in the course of comparing results from different social networks, an analysis was conducted to determine whether any psychometric could correlate equally well in different media environments. For this purpose, a correlation matrix was constructed between the features and psychometrics. Then, due to the small sample size of one of the datasets, an experiment was conducted showing changes in the quality of predictive models, when the initial sample is expanded by adding data from another dataset that has a similar distribution. That is, the possibility of cross-media learning was investigated.