Personality Assessment Based on Multimodal Attention Network Learning With Category-Based Mean Square Error

Area of Science:

• Computer Science
• Psychology
• Artificial Intelligence

Background:

• Personality analysis is crucial for occupational and educational assessments.
• Traditional personality tests can be burdensome due to extensive questionnaires.
• Existing methods may lack nuanced behavioral insights.

Purpose of the Study:

• To develop a more efficient and accurate personality assessment method.
• To leverage multimodal data from daily videos for personality trait determination.
• To reduce the burden on individuals undergoing personality evaluations.

Main Methods:

• A multimodal attention network was developed for personality assessment.
• Utilized daily videos to extract behavioral information: gaze distribution, speech features, facial expressions.
• Introduced a novel attention mechanism based on the facial Region of No Interest (RoNI).
• Employed Category-based Mean Square Error (CBMSE) loss function for improved boundary data distinction.

Main Results:

• The proposed method achieved an average prediction accuracy of 92.07%.
• The RoNI-based attention mechanism enhanced accuracy and reduced network parameters.
• CBMSE effectively distinguished ambiguous personality data points.
• Performance surpassed existing state-of-the-art models on the ECCV 2016 dataset.

Conclusions:

• Multimodal video analysis offers a promising alternative for personality assessment.
• The developed AI model provides a highly accurate and less burdensome approach.
• Attention mechanisms and specialized loss functions significantly improve personality prediction.