Self-Esteem Assessment Based on Self-Introduction: A Multimodal Approach to Personality Computing

Area of Science:

• Psychology
• Computer Science
• Artificial Intelligence

Background:

• Self-esteem is a crucial psychological construct.
• Traditional self-esteem assessment relies on self-report measures like the Rosenberg Self-Esteem Scale (RSES).
• Automated evaluation methods could offer objective and low-burden alternatives.

Purpose of the Study:

• To develop and validate a multimodal self-esteem recognition method using self-introduction videos.
• To achieve automated self-esteem evaluation.
• To assess the generalizability and validity of the developed method.

Main Methods:

• Recruited undergraduate students (N=211 and N=63) for two independent samples.
• Collected 40-second self-introduction videos and RSES scores.
• Extracted visual, audio, and text features; trained late-fusion multimodal models.

Main Results:

• The multimodal model achieved an accuracy of 0.447 and Macro-F1 of 0.438 on the primary dataset.
• The model demonstrated cross-sample generalizability (ACC=0.381, Macro-F1=0.379).
• The method showed good reliability (κ=0.723, ICC=0.745) and correlated with various psychological measures, offering incremental validity for positive affect.

Conclusions:

• Multimodal behavioral features from self-introductions can aid automated self-esteem evaluation.
• This method offers a feasible, low-burden complement to traditional self-report questionnaires.
• Further research can explore its application in diverse populations and contexts.