Why is MBTI personality detection from texts a difficult task?

Abstract

Automatic detection of the four MBTI personality dimensions from texts has recently attracted noticeable attention from the natural language processing and computational linguistic communities. Despite the large collections of Twitter data for training, the best systems rarely even outperform the majority-class baseline. In this paper, we discuss the theoretical reasons for such low results and present the insights from an annotation study that further shed the light on this issue.