A DLF multi-scale quantitative research method for the big five personality traits

This study proposes a DLF Multi-Scale Network Model to analyze personality traits by integrating visual and audio features from video data. Existing methods often overlook the stability and correlation of multimodal features during extraction and fusion, limiting model generalization and recognition accuracy. To address this, we propose the DWTC module, which applies Discrete Wavelet Transform (DWT) to decompose video frames into multi-scale representations. These representations capture hierarchical visual information (e.g., facial expressions, body movements) at varying levels of detail. A Convolutional Neural Network (CNN) is used to extract deeper semantic features from the decomposed representations, ensuring stability and discriminability of visual features. For audio, a multilayer perceptron (MLP) is used for processing, followed by latent feature sequence alignment (LMSA) and integration to address spatio-temporal inconsistencies. This alignment enhances feature complementarity and inter-modal correlations, with final fusion at the decision level. Experiments on the First Impression V1 dataset demonstrate the superior performance of the DLF model in recognizing Big Five personality traits, achieving an average multimodal fusion accuracy of 0.9177, significantly outperforming existing models. This research provides insights into mapping behavioral characteristics to personality traits in digital environments and holds practical value for mental health assessment and personalized recommendation systems.