Scholarlink

With the rapid growth of social media users, detecting depression from vlog (video blog) data has become an emerging research area. However, extracting discriminative multimodal features and efficiently integrating them remain challenging. This paper proposes the Multi-Stage Joint Attention Fusion Network (MSJAFN), which comprises three modules: Channel-Adaptive Focus TCN (CAF-TCN) for adaptive temporal feature extraction, Progressive Multi-Attention Fusion (PMAF) for early cross-modal correlation modeling, and Modality-Decoupled Multi-Scale Deep Fusion (M2DF) for deep multimodal integration. Evaluated on the D-Vlog dataset, MSJAFN achieved an accuracy of 73.13%, a recall of 79.67%, and an F1-score of 76.26%, outperforming most existing models.

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

MSJAFN: Multi-stage Joint Attention Fusion Network for Depression Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yu Yao,
  • Feng Liu,
  • Yuheng Liang,
  • Mingzhou Liu

摘要

With the rapid growth of social media users, detecting depression from vlog (video blog) data has become an emerging research area. However, extracting discriminative multimodal features and efficiently integrating them remain challenging. This paper proposes the Multi-Stage Joint Attention Fusion Network (MSJAFN), which comprises three modules: Channel-Adaptive Focus TCN (CAF-TCN) for adaptive temporal feature extraction, Progressive Multi-Attention Fusion (PMAF) for early cross-modal correlation modeling, and Modality-Decoupled Multi-Scale Deep Fusion (M2DF) for deep multimodal integration. Evaluated on the D-Vlog dataset, MSJAFN achieved an accuracy of 73.13%, a recall of 79.67%, and an F1-score of 76.26%, outperforming most existing models.