Scholarlink

The rise of short-form video platforms like TikTok and Instagram Reels highlights the need for effective background music selection to enhance viewer engagement. We propose MoodFlow, a cross-modal music recommendation system leveraging deep learning to align video emotional tone with mood-appropriate soundtracks. Using Neural ODEs and CNFs, MoodFlow models non-linear transformations between visual and auditory embeddings. A diversity-enhancing soft clustering mechanism balances relevance and genre variety. Experiments show query processing times of 2–5 seconds on GPU hardware, with qualitative assessments confirming high emotional resonance. MoodFlow offers a scalable, mood-aware solution for automated content creation, enhancing user experiences on video-sharing platforms.

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

MoodFlow: Cross-Modal Music Recommendation for Short-Form Videos via Neural ODEs with Mood-Aligned Music Recommendations

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tran Thi Ngan,
  • Pham Duc Minh,
  • Nguyen Duc Quang-Anh,
  • Nguyen Minh-Anh,
  • Nguyen Hai Nam

摘要

The rise of short-form video platforms like TikTok and Instagram Reels highlights the need for effective background music selection to enhance viewer engagement. We propose MoodFlow, a cross-modal music recommendation system leveraging deep learning to align video emotional tone with mood-appropriate soundtracks. Using Neural ODEs and CNFs, MoodFlow models non-linear transformations between visual and auditory embeddings. A diversity-enhancing soft clustering mechanism balances relevance and genre variety. Experiments show query processing times of 2–5 seconds on GPU hardware, with qualitative assessments confirming high emotional resonance. MoodFlow offers a scalable, mood-aware solution for automated content creation, enhancing user experiences on video-sharing platforms.