Scholarlink

<p>Short video platforms have rapidly gained popularity, leading to increased consumption of psychological and self-help videos. However, these videos often present biased, misleading, or inaccurate psychological information that can negatively influence viewers. Existing research mainly concentrates on unimodal sentiment analysis, emotion recognition, or user engagement prediction, while limited attention has been devoted to integrated multimodal frameworks for detecting psychological value transmission bias across textual, audio, and visual modalities. To address this research gap, the study proposes a deep learning-based multimodal architecture for psychological bias detection in short-video environments. The proposed framework employs Bidirectional Encoder Representations from Transformers (BERT) to generate contextual textual embeddings, Waveform to Vector 2.0 for extracting audio representations, and Convolutional Neural Networks (CNNs) for visual feature analysis. Extracted multimodal features are combined through a concatenation-based fusion mechanism and subsequently processed using a Multilayer Perceptron (MLP) classifier for final prediction. The proposed method is evaluated using the Carnegie Mellon University Multimodal Opinion Sentiment and Emotion Intensity (CMU-MOSEI) dataset containing more than 23,000 annotated video clips, from which 2990 positive, neutral, and biased samples were selected. Experimental results demonstrate that the proposed model achieved an accuracy of 0.9624, precision of 0.9631, recall of 0.9624, F1-score of 0.9627, and ROC-AUC of 0.9912. Comparative and ablation studies further confirm that multimodal feature fusion significantly outperforms unimodal approaches in capturing complex psychological patterns and improving large-scale reliability assessment of digital psychological content. Overall, the framework provides an effective scalable solution for monitoring psychological bias transmission across rapidly expanding digital media platforms globally.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluation of the value transmission bias of short video psychological content based on multimodal perception

  • 期刊论文

  • 发表日期：

    2026年7月11日

查看全文
  • Yanghejia Wang

摘要

Short video platforms have rapidly gained popularity, leading to increased consumption of psychological and self-help videos. However, these videos often present biased, misleading, or inaccurate psychological information that can negatively influence viewers. Existing research mainly concentrates on unimodal sentiment analysis, emotion recognition, or user engagement prediction, while limited attention has been devoted to integrated multimodal frameworks for detecting psychological value transmission bias across textual, audio, and visual modalities. To address this research gap, the study proposes a deep learning-based multimodal architecture for psychological bias detection in short-video environments. The proposed framework employs Bidirectional Encoder Representations from Transformers (BERT) to generate contextual textual embeddings, Waveform to Vector 2.0 for extracting audio representations, and Convolutional Neural Networks (CNNs) for visual feature analysis. Extracted multimodal features are combined through a concatenation-based fusion mechanism and subsequently processed using a Multilayer Perceptron (MLP) classifier for final prediction. The proposed method is evaluated using the Carnegie Mellon University Multimodal Opinion Sentiment and Emotion Intensity (CMU-MOSEI) dataset containing more than 23,000 annotated video clips, from which 2990 positive, neutral, and biased samples were selected. Experimental results demonstrate that the proposed model achieved an accuracy of 0.9624, precision of 0.9631, recall of 0.9624, F1-score of 0.9627, and ROC-AUC of 0.9912. Comparative and ablation studies further confirm that multimodal feature fusion significantly outperforms unimodal approaches in capturing complex psychological patterns and improving large-scale reliability assessment of digital psychological content. Overall, the framework provides an effective scalable solution for monitoring psychological bias transmission across rapidly expanding digital media platforms globally.