Answers are grounded in this paper and cite their sources.
In view of the semantic gap problem existing in the existing multimodal video emotion analysis model due to the insufficient fusion process of different modes, which affects the performance of emotion analysis, a multimodal video emotion analysis model based on multi-head attention mechanism and multimodal cross fusion is proposed. The model first extracts the single modal features of text, voice, vision (image) and so on in the video; Then, the GRU network is used to extract the temporal characteristics of each single mode context; Then, the multi-head attention mechanism is used to fuse text-voice, text-video and voice-video, and then the single-mode features and the features after the fusion are cross-mode fused; Finally, through the attention mechanism, the fused features are input into the emotion classification network for classification. The experimental results show that compared with the existing models, this model can better fuse the attribute features within and between modes, and has a certain improvement in the accuracy of emotion recognition and F1 value.
No takes yet. Share an insight, caveat, or question.
Zhao et al. (2023) studied this question.