How to Obtain Reliable Labels for MBTI Classification from Texts?

Proceedings of Recent Advances in Natural Language Processing, pages 1360–1368 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_152 1360 How to Obtain Reliable Labels for MBTI Classification from Texts? Sanja ˇStajner Symanto Research Nuremberg, Germany sanja.stajner@symanto.com Seren Yenikent Symanto Research Nuremberg, Germany seren.yenikent@symanto.com Abstract Automatic detection of the Myers-Briggs Type Indicator (MBTI) from short posts attracted noticeable attention in the last few years. Re- cent studies showed that this is quite a diffi- cult task, especially on commonly used Twitter data. Obtaining MBTI labels is also difficult, as human annotation requires trained psychol- ogists, and automatic way of obtaining them is through long questionnaires of questionable usability for the task. In this paper, we present a method for collecting reliable MBTI labels via only four carefully selected questions that can be applied to any type of textual data. 1 Introduction The Myers-Briggs Type Indicator (MBTI) model (Briggs-Myers and Myers, 1995) is one of the most widely used non-clinical psychometric models (ˇStajner and Yenikent, 2020). It classifies people into two groups across four dimensions: extraver- sion/introversion (E/I), sensing/intuition (S/N), thinking/feeling (T/F), and judgement/perception (J/P). This leads to a total of 16 personality types. The first three dimensions are based on the theo- retical work of Carl Jung (1921), while the fourth dimension was added later by Myers and Briggs- Myers (1995). The MBTI personality framework has already been used for decades in educational and industry settings, e.g. for finding jobs that best resonate with the person’s preferences for informa- tion processing (S/N and T/F dimensions), finding work organization types that best resonate with the person’s preferred judgement processes (J/P dimen- sion) thus leading to better job satisfaction, and for better matching work environments with the person’s preferences (E/I dimension) to lower em- ployee turnover (Briggs-Myers and Myers, 1995). The original MBTI questionnaire contains 93 questions and is not freely available.1 Due to the 1https://www.myersbriggs.org/using-type-as-a- popularity of MBTI framework (it is estimated that more than 2 million US adults complete the inven- tory every year),2 there is a number of freely avail- able alternative MBTI questionnaires on the inter- net, with the 16personalities test3 being one of the most popular ones. According to the Myers-Briggs Foundation4 and the 16personality test website,5 both questionnaires satisfy the accepted standards for test validity and reliability. Nevertheless, the MBTI questionnaires have received a noticeable criticism from the academic community (Pittenger, 1993; Boyle, 1995) for not relying on a scientif- ically proven (i.e. data-driven) background, but rather on qualitative measures such as observation and introspection. The other common criticism is the binary nature of the questionnaire as it is known that the majority of people usually lies somewhere in the middle of the scales (Pittenger, 1993). The questionnaire-based personality detection has several weaknesses: it requires trained human assessors; it is prone to social desirability bias (Krumpal, 2011) and reference-group effect (Heine et al., 2002); it is questionable if answering ques- tionnaires is a natural way of showing ones per- sonality (as opposed to free writing or behaviour “when nobody watches”). To …

Content truncated to 500 words