Combining machine learning algorithms for personality trait prediction

Abstract

Personality is a unique trait that allows discriminating between individuals. It can be defined by a set of stable characteristics of an individual that may affect their interactions, relationships, attitudes, behaviors, and even psychological health. Currently, with the advent of social networking sites that provide user-generated text content, personality trait recognition has gained a lot attention. These texts from social networks keep a record of users' psychological activity over time, which makes it a vital piece of information to analyze the users' personality traits. This study proposes a stacked ensemble model combining multiple classic machine learning classifiers using different semantic and lexical features, as well as deep learning algorithms, and distinct word embedding techniques to develop a personality recognition model. The performance of the proposed ensemble model has been assessed using the gold standard MyPersonality dataset. The results demonstrate that the proposed framework outperforms different ensemble model architectures, classical machine learning, and deep learning-based algorithms, as well as state-of-the-art studies, achieving an average accuracy of 72.69%.

Keywords

• Personality trait detection
• Stacked ensemble
• Big five model
• Feature extraction
• Word embedding