Short video traffic prediction method based on multi-modal data
By constructing a multimodal short video traffic prediction model, the problems of insufficient dynamic temporal evolution of multimodal features and cross-modal correlation in short video traffic prediction are solved, achieving more accurate and stable traffic prediction and supporting content recommendation and advertising placement decisions.
Patent Information
- Application Number
- CN202511093352.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-06
- Publication Date
- 2025-10-28
AI Technical Summary
Existing short video traffic prediction methods struggle to effectively capture the dynamic temporal evolution of multimodal features, fail to fully explore the potential correlations between video, audio, and text modalities, and the text information cannot be correlated with frame-level video/audio, leading to prediction bias and noise.
A multimodal short video traffic prediction model is constructed, including a multimodal feature extraction module, a temporal modeling module, a modal attention module, a global modulation module, and a prediction head module. Through multimodal feature extraction, temporal modeling, modal attention fusion, and adaptive loss optimization, cross-modal interaction modeling and information fusion are achieved.
It significantly improves the accuracy and stability of short video traffic prediction, reduces reliance on manual feature engineering, and provides a reliable basis for content recommendation and advertising.