Research on Short Video Data Analysis Based on Multimodal Features

As an emerging media format, short videos have become an important means for people to obtain information and entertainment. However, the complexity and diversity of short video content pose significant challenges for data analysis. This paper provides a reference for short video data analysis research by investigating multimodal data feature extraction, emotion recognition, and user interest prediction. Firstly, this study explores multimodal data feature extraction methods, utilizing deep learning models to extract image, audio, and textual features. Secondly, an emotion recognition method based on a user feature-guided attention mechanism is proposed, which enhances the feature representation of emotional analysis by fusing multimodal features. Finally, a user interest prediction model is designed by integrating multimodal features and user interest evolution patterns.