Mathematical representations of MBTI personality type and their impact on predictive performance: An application in career counseling

Summary

Personalization plays a very important role in modern society and many companies create personalized products and services based on costumer profiles. Personalized services can be further improved through the adoption of personality theory, which describes individual behaviour and motivation. Recent developments in the domain of automated personality computing allowed for personality inference from user social media data making personality profiling of larger audiences scalable. This led to the emergence of digital ecosystems, such as personality-aware recommendation systems and personality-based content generation platforms. These systems represent human personality through appropriate measurable quantities derived from two popular personality models: Big5 and MBTI. MBTI data, when used for predictive tasks, tend to result to lower performance than Big5 data. However, MBTI is more popular and, thus, there are more data available.

The present thesis aims to increase predictive performance of MBTI-related tasks through exploration of potential mathematical representations of MBTI personality. The research focused on KPMI, a data set from the career counseling sector that was used to train different machine learning algorithms to predict employee job satisfaction. The experiments were repeated for various mathematical representations of MBTI to evaluate their performance. Results were mainly inconclusive, although there is some evidence that the raw test scores provided in KPMI, before being transformed to categorical MBTI types through thresholding, tend to perform better. The inconclusive results are attributed to lack of information in the data set to tackle the problem at hand. Consequently, it is suggested to enrich the data set with external information and reapply the proposed evaluation framework. Alternatively, the evaluation framework can be applied to other data sets and predictive tasks.