Personality Predictive Lexical Cues and Their Correlations

Proceedings of Recent Advances in Natural Language Processing, pages 514–523 Sep 1–3, 2021. https://doi.org/10.26615/978-954-452-072-4_058 514 Personality Predictive Lexical Cues and Their Correlations Xiaoli He Rutgers University New Brunswick, NJ, USA hxl.daybreak@gmail.com Gerard de Melo Hasso Plattner Institute / University of Potsdam Potsdam, Germany http://gerard.demelo.org Abstract In recent years, a number of studies have used linear models for personality prediction based on text. In this paper, we empirically analyze and compare the lexical signals captured in such models. We identify lexical cues for each dimension of the MBTI personality scheme in several different ways, considering different datasets, feature sets, and learning algorithms. We conduct a series of correlation analyses be- tween the resulting MBTI data and explore their connection to other signals, such as for Big-5 traits, emotion, sentiment, age, and gen- der. The analysis shows intriguing correlation patterns between different personality dimen- sions and other traits, and also provides evi- dence for the robustness of the data. 1 Introduction The notion of personality refers to an individual’s characteristic patterns of thinking, feeling, and be- having (Sherman et al., 2013). Studies have shown that personality influences an individual’s language usage (Schwartz et al., 2013b; Tucker, 1968; Hirsh and Peterson, 2009). Hence, language may reveal subtle cues about an individual’s personality. Since an individual’s personality is known to be fairly stable across long periods of time, the relation be- tween personality and language usage is expected to be analyzable given sufficiently large amounts of textual data. Motivation. Many people now routinely post in- formation pertaining to their daily life, thoughts, emotions, and opinions on different online social media platforms. A number of studies have shown that such social media text may enable automated personality predictions, as reviewed in more detail in Section 2. Yet, automatic personality prediction is still a challenging problem, and there remain several unresolved issues. First, due to privacy concerns and the high la- beling cost, the number of publicly available la- beled datasets is limited, and the sample size in such datasets is often rather small (especially when compared with the high dimensionality of n-gram features). Beyond this, some datasets only provide a small number of sentences per sample. These limitations make it difficult to know to what ex- tent results in individual studies generalize across different datasets. Second, it is non-trivial to compare results across different studies, as they adopt different feature representations and machine learning methods, and consider different personality models. Two particularly well-known personality mod- els are Myers-Briggs Type Indicators (MBTI) and Big-5 traits, which we introduce in more detail in Section 2. In the field of personality psychol- ogy, studies have shown clear correlations be- tween self-reported MBTI and Big-5. For instance, MBTI’s INTROVERSION–EXTRAVERSION corre- lates with Big-5 EXTRAVERSION, MBTI’s SENS- ING–INTUITION and JUDGING–PERCEIVING cor- relate with Big-5’s OPENNESS trait, and MBTI’s JUDGING–PERCEIVING also correlates with Big-5 CONSCIOUSNESS (Tobacyk et al., 2008). This raises the question of whether signals from naturally occurring text exhibit similar connections, and how they relate to other psychological and …

Content truncated to 500 words