DLP-personality detection: a text-based personality detection framework with psycholinguistic features and pre-trained features

Hao Lin

Abstract: Text-based personality detection aims to identify the personality traits implied in subject-supplied textual data. However, over-reliance on pre-trained language models and neglect of psycholinguistic features has become a bottleneck in personality detection. In this work, we conduct extensive feature-level ablation experiments using multiple psycholinguistic features to verify the importance of psycholinguistic features for personality detection. Furthermore, we propose a novel personality detection framework, DLP-Personality Detection, which fuses multiple psycholinguistic features and pre-trained language features. With the DLP-Personality Detection, we achieve state-of-the-art performance for the Big Five personality traits (Big 5) and Myers-Briggs Type Indicator (MBTI) personality traits on the Essays and Kaggle MBTI datasets. Finally, we provide some suggestions for psycholinguistic features and discuss future work for personality detection.