TwiSty: A Multilingual Twitter Stylometry Corpus for Gender and Personality Profiling

TWISTY: a Multilingual Twitter Stylometry Corpus for Gender and Personality Profiling Ben Verhoeven,† Walter Daelemans† and Barbara Plank∗ †CLiPS Research Center, University of Antwerp, Belgium ∗University of Groningen, The Netherlands {ben.verhoeven,walter.daelemans}@uantwerpen.be,b.plank@rug.nl Abstract Personality profiling is the task of detecting personality traits of authors based on writing style. Several personality typologies exist, however, the Myers-Briggs Type Indicator (MBTI) is particularly popular in the non-scientific community, and many people use it to analyse their own personality and talk about the results online. Therefore, large amounts of self-assessed data on MBTI are readily available on social-media platforms such as Twitter. We present a novel corpus of tweets annotated with the MBTI personality type and gender of their author for six Western European languages (Dutch, German, French, Italian, Portuguese and Spanish). We outline the corpus creation and annotation, show statistics of the obtained data distributions and present first baselines on Myers-Briggs personality profiling and gender prediction for all six languages. Keywords: computational stylometry, author profiling, multilingual corpus 1. Introduction Personality prediction is one of the most difficult author profiling tasks in computational stylometry. It involves de- tecting personality traits on the basis of writing style. Sev- eral typologies of personality traits exist, but the two most well-known are Big Five (Goldberg, 1990) and Myers- Briggs Type Indicator (Briggs Myers and Myers, 2010). Modeling author attributes such as personality plays an in- creasingly important role in numerous applications, from business intelligence to personalized translation (Mirkin et al., 2015). A large body of work exists on predicting in- dividual author attributes from linguistic input (Rosenthal and McKeown, 2011; Nguyen et al., 2011; Eisenstein et al., 2011; Volkova et al., 2013; Alowibdi et al., 2013; Ciot et al., 2013; Volkova et al., 2015). Apart from demographic features, such as age or gender, there is also a growing inter- est in predicting psychological properties such as personal- ity, attested by a growing literature and recent shared tasks on this topic (Celli et al., 2013; Celli et al., 2014; Rangel et al., 2015). Predicting personality types is not only of interest for psy- chology, but also for commercial purposes to even health care. Recent work by Preot¸iuc-Pietro et al. (2015) investi- gated the link between personality types, social media be- havior, and psychological disorders, such as depression and post-traumatic stress disorder. They found that certain per- sonality traits are predictive of mental illness. However, computational personality recognition is ham- pered by the availability of limited amounts of labeled data (Nowson and Gill, 2014). Many early existing data sets contain written essays of a certain topic, which are written in highly canonical language. Such controlled set- tings inhibit the expression of individual traits much more than spontaneous language. As such data is hard to obtain, only limited amounts were available. With the availability of social media text, recent efforts shifted toward using such data (Schwartz et al., 2013a; Schwartz et al., 2013b; Park et al., 2015; Kosinski et al., 2015). For example, Kosinski et al. (2015) collected a large amount …

Content truncated to 500 words