We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
MC-CA:基于模态时序列耦合与交互式多头注意力的多模态情感分析.
- Authors
张涛; 郭青冰; 李祖贺; 邓璐娟
- Abstract
Multimodal sentiment analysis aims to comprehensively analyze multimodal data to obtain people's opinions and attitudes. However, the existing models ignore the problem of multimodal interaction, which leads to their limited performance. Aiming at this problem, this paper proposes a multimodal sentiment analysis model (MC-CA) based on modal temporal coupling and interactive multi-head attention. Firstly, the model utilizes affine transformation to fuse the sentiment information and the timing information. Then the model applies the interactive multi-head attention mechanism to obtain the interaction information among modalities. Finally, the model employs a multi-channel sentiment prediction method to integrate global information and local information to achieve multimodal collaborative training. Experimental results on multiple public datasets show that this model can establish interactions between multimodal data and achieve excellent performance in multimodal sentiment analysis tasks.
- Subjects
SENTIMENT analysis; AFFINE transformations; TASK analysis; ATTITUDE (Psychology); FORECASTING; MODAL logic
- Publication
Journal of Chongqing University of Posts & Telecommunications (Natural Science Edition), 2023, Vol 35, Issue 4, p680
- ISSN
1673-825X
- Publication type
Article
- DOI
10.3979/j.issn.1673-825X.202205090107