Personality Recognition from Facebook Text

This work concerns a study in the Natural Language Processing field aiming to recognise personality traits in Portuguese written text. To this end, we first built a corpus of Facebook status updates labelled with the personality traits of their authors, from which we trained a number of computational models of personality recognition. The models include a range of alternatives ranging from a standard approach relying on lexical knowledge from the LIWC dictionary and others, to purely text-based methods such as bag of words, word embeddings and others. Results suggest that word embedding models slightly outperform the alternatives under consideration, with the advantage of not requiring any language-specific lexical resources.